Circuit and method for generating filler pixels from the original pixels in a video stream

[0001] This is a continuation-in-part of co-pending International Application PCT/US/99/17606 filed on Aug. 3, 1999 designating the United States, which claims the benefit of provisional application 60/095,201 filed on Aug. 3, 1998.

[0002] The invention relates generally to electronic circuits, and more particularly to a circuit and method for estimating the values of filler pixels from the values of original pixels in a video stream. For example, one can use the circuit or method to de-interlace an interlaced video stream. That is, from the original pixels in the original video fields, one can generate the values of filler pixels, form complementary filler video fields from the filler pixels, and combine the original and complementary fields to generate respective video frames.

[0003] FIG. 1 is a diagram of an interlaced video frame 10, which includes an even video field 12 and an odd video field 14. The even field 12 includes the even lines a0, a2, a4 . . . a(n-1) of the frame 10, and the odd field 14 includes the odd lines b1, b3, b5 . . . bn of the frame 10. A video source (not shown) such as a video camera generates the even field 12 at a time t₀ and generates the odd field 14 at a subsequent time t₁, and a video display (not shown in FIG. 1) displays the fields 12 and 14 in the same sequence. For example, according to the National Television Standards Committee (NTSC) video standard, which has been in existence for over 50 years, a video source generates and a video display displays one field 12 or 14 every {fraction (1/60)}^th of a second, and thus respectively generates and displays one frame 10 every {fraction (1/30)}^th of a second. But even though the display displays the fields 12 and 14 at different times, the relatively slow frequency responses of the display and the human eye cause a viewer to perceive that the display is displaying the fields 12 and 14 simultaneously. Thus, the viewer perceives the frame 10 as a single image instead of two sequential partial images.

[0004] Many modern video applications, however, generate streams of non-interlaced, i.e., progressive, video frames. For example, most applications of the new High-Definition Television (HDTV) standards such as MPEG (Motion Pictures Experts Group) call for the generation and display of an entire frame 10 approximately every {fraction (1/60)}^th of a second. Because such MPEG video sources and displays respectively generate and display all the lines of a progressive frame at one time and not at two sequential times, progressive frames contain little if any motion blurring.

[0005] Because many existing video sources generate interlaced video, and because many existing video works are recorded in an interlaced format, one may wish to convert a stream of interlaced video frames into a stream of progressive video frames that are compatible with HDTV systems. For example, one may wish to convert a VHS signal from a VCR (not shown) into a progressive video signal for display on an HDTV display (not shown in FIG. 1).

[0006] Still referring to FIG. 1, a simple technique for de-interlacing the interlaced frame 10 is to merge the fields 12 and 14 into a resulting progressive frame that is displayed twice in a row at the frame-display rate. For example, in the MPEG standard described above, a display displays this resulting progressive frame and then displays it again {fraction (1/60)}^th of a second later. But because the fields 12 and 14 were generated at different times t₀ and t₁, the resulting progressive frame may contain blurred regions, particularly if there were changes in the image contents, i.e., motion, between the times t₀ and t₁. Thus unfortunately, this technique often results in a video stream of relatively poor visual quality by HDTV standards.

[0007] Another technique for de-interlacing the video frame 10 is to generate respective complimentary filler fields for the original fields 12 and 14. That is, for the even field 12, one “fills” the missing odd lines with odd filler lines, and for the odd field 14, one fills the missing even lines with even filler lines.

[0008] One approach to this filler technique is to spatially interpolate the filler pixels of the filler lines from the values of neighboring original pixels within the same field. This approach is typically most accurate when there is significant motion between the original fields. Unfortunately, many spatial interpolation approaches have a tendency to falsely interpolate a thin line as a directional edge, and thus introduce artifacts into the resulting progressive frame.

[0009] An alternative approach is to temporally interpolate the values of the filler pixels from the values of corresponding original pixels in adjacent complimentary original fields. This approach is typically most accurate when there is little or no motion between the original fields.

[0010] Because many interlaced video streams have some segments that exhibit significant inter-field motion and other segments that exhibit little or no inter-field motion, another approach, often called a hybrid approach, combines the spatial and temporal interpolation approaches. For example, one hybrid approach varies the relative weightings of the temporal and spatial interpolation approaches based on the magnitude of inter-field motion. The greater the magnitude of inter-field motion, the more heavily weighted the spatial interpolation approach; conversely, the lower the magnitude of inter-field motion, the more heavily weighted the temporal interpolation approach.

[0011] Unfortunately, many hybrid techniques sometimes fail to detect significant inter-field motion, and thus assign improper weightings to the spatial and temporal approaches. Although some of these techniques can be modified to overcome this defect, such modifications often require an impractical amount of memory.

[0012] To help the reader more easily understand the concepts discussed below in the description of the invention, the following is a basic overview of the relevant aspects of conventional image-compression techniques.

[0013] To electronically transmit a relatively high-resolution image over a relatively low-band-width channel, or to electronically store such an image in a relatively small memory space, it is often necessary to compress the digital data that represents the image. Such image compression typically involves reducing the number of data bits necessary to represent an image. For example, High-Definition-Television (HDTV) video images are compressed to allow their transmission over existing television channels. Without compression, HDTV video images would require transmission channels having bandwidths much greater than the bandwidths of existing television channels. Or, to reduce data traffic and transmission time to acceptable levels, one may compress an image before sending it over the internet. In addition, to increase the image-storage capacity of a CD-ROM or server, on may compress an image before storing it.

[0014] Referring to FIGS. 2A-3D, the basics of the popular block-based MPEG compression standards, which include MPEG-1 and MPEG-2, are discussed. FIGS. 2A-2D illustrate compressing a Y-C_B-C_R image (e.g., video frames or fields) according to an MPEG 4:2:0 format, and FIGS. 3A-3D illustrate compressing a Y-C_B-C_R image according to an MPEG 4:2:2 format. But the discussed concepts also apply to other MPEG formats, to images that are represented in other color spaces, and to other block-based compression standards such as the Joint Photographic Experts Group (JPEG) standard, which is often used to compress still images. Although many details of the MPEG standards and the Y, C_B, C_R color space are omitted for brevity, these details are well-known and are disclosed in a large number of available references including “Video Compression” by Peter D. Symes, McGraw-Hill, 1998, which is incorporated by reference. Furthermore, other well-known block-based compression techniques are available for encoding and decoding video and still images.

[0015] Referring to FIGS. 2A-2D, the MPEG standards are often used to compress temporal sequences of images—video frames for purposes of this discussion—such as found in a television broadcast. Each video frame is divided into subregions called macro blocks, which each include one or more pixels. FIG. 2A is a 16-pixel-by-16-pixel macro block 20 having 256 pixels 22 (not drawn to scale). In the MPEG standards, a macro block is always 16×16 pixels, although other compression standards may use macro blocks having other dimensions. In the original video frame, i.e., the frame before compression, each pixel 22 has a respective luminance value Y and a respective pair of color-, i.e., chroma-, difference values C_B and C_R.

[0016] Before compression of the video frame, the digital luminance (Y) and chroma-difference (C_B and C_R) values that will be used for compression, i.e., the original or pre-compression values, are generated from the original Y, C_B, and C_R values of the original frame. In the MPEG 4:2:0 format, the pre-compression Y values are the same as the original Y values. Thus, each pixel 22 merely retains its original luminance value Y. But to reduce the amount of data to be compressed, the MPEG 4:2:0 format allows only one pre-compression C_B value and one pre-compression C_R value for each group 24 of four pixels 22. Each of these pre-compression C_B and C_R values are respectively derived from the original C_B and C_R values of the four pixels 22 in the respective group 24. For example, a pre-compression C_B value may equal the average of the original C_B values of the four pixels 22 in the respective group 24. Thus, referring to FIGS. 2B-2D, the pre-compression Y, C_B, and C_R values generated for the macro block 20 are arranged as one 16×16 matrix 26 of pre-compression Y values (equal to the original Y values for each respective pixel 22), one 8×8 matrix 28 of pre-compression C_B values (equal to one derived C_B value for each group 24 of four pixels 22), and one 8×8 matrix 30 of pre-compression C_R values (equal to one derived C_R value for each group 24 of four pixels 22). The matrices 26, 28, and 30 are often called “blocks” of values. Furthermore, because it is convenient to perform the compression transforms on 8×8 blocks of pixel values instead of on 16×16 blocks, the block 26 of pre-compression Y values is subdivided into four 8×8 blocks 32a-32d, which respectively correspond to the 8×8 blocks A-D of pixels 22 in the macro block 20. Thus, referring to FIGS. 2A-2D, six 8×8 blocks of pre-compression pixel data are generated for each macro block 20; four 8×8 blocks 32a-32d of pre-compression Y values, one 8×8 block 28 of pre-compression C_B values, and one 8×8 block 30 of pre-compression C_R values.

[0017] FIGS. 3A-3D illustrate the generation of the pre-compression Y, C_B, C_R, values according to the MPEG 4:2:2 format. Referring to FIG. 3A, the pixels 22 of the macro block 20 are arranged in two-pixel groups 34 as compared to the four-pixel groups 24 (FIG. 2A) that the 4:2:0 format calls for. Referring to FIG. 3B, in the MPEG 4:2:2 format, the pre-compression Y values are the same as the original Y values. Thus, as in the 4:2:0 format, each pixel 22 merely retains its original luminance value Y. But referring to FIGS. 3C and 3D, to reduce the amount of data to be compressed, the MPEG 4:2:2 format allows only one pre-compression C_B value and one pre-compression C_R value for each group 34 of two pixels 22. Each of these pre-compression C_B and C_R values are respectively derived from the original C_B and C_R values of the two pixels 22 in the respective group 34. For example, a pre-compression C_B value may equal the average of the original C_B values of the two pixels 22 in the respective group 34. Therefore the 4:2:2 format calls for twice as many C_B and C_R values (one per every two original pixels) as the 4:2:0 format (one per every four pixels). Thus, referring to FIGS. 3B-3D, the pre-compression Y. C_B, and C_R values generated for the macro block 20 of FIG. 3A are arranged as one 16×16 matrix 36 of pre-compression Y values (equal to the original Y values for each respective pixel 22), one 8×16 matrix 38 of pre-compression C_B values (equal to one derived C_B value for each group 34 of two pixels 22), and one 8×16 matrix 40 of pre-compression C_R values (equal to one derived C_R value for each group 34 of two pixels 22). As discussed above, because it is convenient to perform the compression transforms on 8×8 blocks of pixel values instead of 16×16 or 8×16 blocks, the block 36 of pre-compression Y values is subdivided into four 8×8 blocks 42a-42d, which respectively correspond to the 8×8 blocks A-D of pixels in the macro block 20. Likewise, the block 38 of pre-compression C_B values is subdivided into two 8×8 blocks 44a and 44b, which correspond to the pairs of blocks A and B and C and D, respectively. Similarly, the block 40 of pre-compression C_R values is subdivided into two 8×8 blocks 46a and 46b, which correspond to the pairs of blocks A and B and C and D, respectively. Thus, referring to FIGS. 3A-3D, eight 8×8 blocks of pre-compression pixel data are generated for each macro block 20: four 8×8 blocks 42a-42d of pre-compression Y values, two 8×8 blocks 44a-44b of pre-compression C_B values, and two 8×8 blocks 46a-46a of pre-compression C_R values.

[0018] In one aspect of the invention, an image processing circuit includes a processor that receives a value of an original pixel of an original first video image and a value of an original pixel of an original second video image. The processor generates a first pixel-value component from the value of the original pixel of the first original video image, and generates a second pixel-value component from the value of the original pixel in the original second video image. From the first and second pixel-value components, the processor generates a value of a filler pixel, and combines the filler pixel and the original first video image to generate a resulting video image.

[0019] One can use such an image processing circuit to generate a filler video field from an original video field and to merge the filler and original fields to generate a resulting video frame. Such an image processing circuit often uses less memory and detects inter-field motion more accurately than prior image processing circuits.

[0020] In another aspect of the invention, the processor of the image processing circuit receives first and second sets of pixel values for first and second respective groups of original pixels in an original video image. The processor calculates direction values from the first and second sets of pixel values for a filler pixel that is for disposition in the original video image between the first and second groups of original pixels. The processor generates a value for the filler pixel based on the calculated direction values.

[0021] One can use such an image processing circuit to spatially interpolate the filler-pixels of a filler field from the original pixels in an original field and to merge the filler and original fields to generate a resulting video frame. Such an image processing circuit often distinguishes thin lines from edges more accurately, and thus often produces fewer visual artifacts, than prior image processing circuits.

[0044] FIG. 4 is a block diagram of an image processing circuit 50 according to an embodiment of the invention. The circuit 50 can calculate the values of filler pixels from the original pixels in the original video fields, generate filler fields from the filler pixels, and merge the filler fields and original fields to generate resulting video frames. In one embodiment, the circuit 50 extends the influence, of detected inter-field motion to more filler fields than prior image processing circuits. Thus, the circuit 50 often generates more accurate filler fields and higher-quality resulting frames than prior circuits. In another embodiment, the circuit 50 uses less memory for storing motion information than many prior circuits. In yet another embodiment, the circuit 50 spatially interpolates thin lines and edges more accurately than many prior circuits, and this further increases the accuracy of the filler fields and the visual quality of the resulting frames.

[0045] The image processing circuit 50 includes a field buffer 52, which receives and stores one or more original video fields from a stream of original video fields. A processor unit 54 includes a processor 56 and a memory 58, generates filler fields from the original fields stored in the buffer 52, and merges the filler and original fields to generate respective resulting video frames. A frame buffer 60 stores the frames generated by the unit 54 and provides them for display on a video display 62. In one embodiment, the processor 56 is a Very Long Instruction Word (VLIW) processor manufactured by Equator Technologies of Seattle, Wash. In another embodiment, the unit 54 generates the resulting frames in a HDTV format and the display 62 is an HDTV display. In yet another embodiment, a high-capacity storage device such as a Digital Video Disk (DVD) replaces the frame buffer 60 and stores the resulting video frames for later display.

[0046] Referring to FIG. 5, the operation of the image processing circuit 50 of FIG. 4 according to an embodiment of the invention is discussed in conjunction with the flow diagram 63.

[0047] Referring to block 64 of the flow diagram 63, the processor unit 54 retrieves from the field buffer 52 the values of the original pixels that respectively compose the next two original non-complimentary video fields in the sequence of original video fields.

[0048] Next, referring to block 66, from the retrieved values of the original pixels the processor 56 calculates a motion value for a group of filler pixels. This group includes one or more filler pixels that the processor 56 will merge with the first of the two non-complimentary original fields to form a resulting frame.

[0049] Referring to block 68, the processor 56 spatially interpolates a respective pixel value for each of the filler pixels in the group.

[0050] Referring to block 70, the processor 56 also temporally interpolates a respective pixel value for each filler pixel in the group.

[0051] Next, referring to block 72, the processor 56 calculates respective spatial and temporal weighting factors from the motion value and weights the spatially and temporally interpolated pixel values with these respective factors. The processor 56 then combines these weighted pixel values to generate a respective resulting pixel value for each filler pixel in the group. The processor unit 54 stores these resulting filler pixel values in the frame buffer 60, or stores them in the memory 58 until the processor 56 generates the entire filler field.

[0052] Referring to block 74, if the processor 56 must generate more filler-pixel values to complete the respective filler field, then the processor unit 54 returns to block 66. Conversely, referring to block 76, if the processor 56 has completed the current filler field but there are more original fields in the buffer 52, then the processor unit 54 returns to block 64. But if there are no more original fields in the field buffer 52, then the processor unit 54 halts filler-field generation.

[0053] Referring to FIGS. 6-18E, the steps of the flow diagram 153 of FIG. 5 are discussed in more detail according to an embodiment of the invention.

[0054] FIG. 6 is a timing diagram of a sequence of video fields that the field buffer 52 of FIG. 4 receives according to an embodiment of the invention. The sequence includes four even fields E₀-E₃, which alternate with four odd fields O₀-O₃ for a total of eight fields, i.e., four interlaced frames. Each field occurs at a relative time t within the sequence. For example, the odd field O₃ is the most recent field at time t₇, and the even field E₀ is the least recent field at time t₀. Furthermore, the original lines of each field are shown as closed blocks, and the filler lines of the complimentary filler fields are shown as open blocks. For example, the original field E₀ includes original even lines a0, a2, a4, . . . , a(k−1), and the odd filler field that the image processing circuit 50 of FIG. 4 will generate for E₀ includes filler odd lines a1, a3, a5, . . . , a(k). Likewise, the original field O₀ includes original odd lines b1, b3, b5, . . . , b(k), and the even filler field that the circuit 50 will generate for O₀ includes filler even lines b0, b2, b4, . . . , b(k−1). Although the filler lines are shown for clarity in explaining the filler-field generation process discussed below, one should understand that the field buffer 52 does not receive the filler lines. Furthermore, in this embodiment, a video frame has an even number of lines. Therefore, because a video field has half the lines of the video frame, each of the original video fields E₀-E₃ and O₀-O₃ also has an even number of lines. Thus, k-i is an even number and k is an odd number because the first line number is 0. As discussed below, the circuit 50 generates the values of the filler pixels from the original pixels in the complimentary field and in other fields of the same polarity. For example, in one embodiment, the circuit 50 generates the values of the filler pixels in the filler line a1 of E₀ from the values of the original pixels in the original lines a0 and a2 of E₀ and the original lines c0 and c2 of E₁.

[0055] Referring to FIG. 7, the generation of motion values for filler pixels in odd filler fields is discussed according to an embodiment of the invention. For example purposes, the generation of motion values is discussed in conjunction with the original even fields E₀ and E₁ being represented in a Y, C_B, and C_R color space and having been compressed according to the MPEG 4:2:0 format, it being understood that the same principles apply to the other original even fields of the FIG. 6 sequence. The generation of motion values for filler pixels in even filler fields is discussed below in conjunction with FIG. 11.

[0056] FIG. 7 is a diagram of the original and filler pixels that compose the original and filler lines, respectively, of the even fields E₀ and E₁. The pixels of the original even field E₀ and its corresponding odd filler field are denoted as P_kx, and the pixels of E₁ and its corresponding odd filler field are denoted as P′_kx, where k denotes the line and x denotes the column. Like the original pixels, the filler pixels are arranged in 2×2 blocks of four pixels (see FIG. 2A). For example, a block 80 includes filler pixels P₁₂, P₁₃, P₃₂, and P₃₃, which compose the complimentary filler field for E₀.

[0057] Still referring to FIG. 7, the image processing circuit 50 of FIG. 4 generates a respective raw motion value RM for each interior block of filler pixels that compose the complimentary filler field for E₀. (The motion analysis of the exterior blocks, i.e., the blocks that include the first two and last two pixels of a line, are discussed below.) The circuit 50 calculates RM from the differences in the luminance and chrominance values of the original pixels that border the filler block in E₀ and the corresponding original pixels in E₁. For example, the luminance difference values for the filler-pixel block 80 are given by the following equation:
DY_ij=|Y_ij−Y′_ij|_{i=0,2; j=1,2,3,4}  1)
Thus, in this embodiment, there are eight luminance difference values DY:|Y₀₁−Y′₀₁|, |Y₀₂−Y′₀₂|, |Y₀₃−Y′₀₃|, |Y₀₄−Y′₀₄|, |Y₂₁−Y′₂₁|, |Y₂₂−Y′₂₂|, |Y₂₃−Y′₂₃|, and |Y₂₄−Y′₂₄|. Here, Y₀₁ is the luminance value for the original pixel P₀₁, Y′₀₁ is the luminance value for P′₀₁, Y₀₂ is luminance value for the original pixel P₀₂, and so on. The C_R difference value is given by the following equation:
DC_R01=|C_R01−C′_R01|  2)
where C_R01 is the C_R value for the block of original pixels in E₀ including P₀₂, P₀₃, P₂₂, and P₂₃, and C′_R01 is the C_R value for the block of original pixels in E₁ including P′₀₂, P′₀₃, P′₂₂, and P′₂₃. Similarly, the C_B difference value is given by the following equation:
DC_B01=|C_B01−C_B01|  3)
where the blocks of original pixels for C_B01 and C′_B01 are the same as the blocks for C_R01 and C′_R01, respectively.

[0058] Still referring to FIG. 7, the image processing circuit 50 of FIG. 5 calculates the raw motion value RM₀₁, for the block 80 as the maximum of the average luminance and chrominance differences according to the following equation: $\begin{array}{cc}\begin{array}{cc}{\mathrm{RM}}_{01}=\mathrm{Max}\left[\frac{1}{8}\sum _{i_{\mathrm{even}}=0}^2\sum _{j=0}^4{\mathrm{DY}}_{\mathrm{ij}},D{C}_{\mathrm{R01}},D{C}_{\mathrm{B01}}\right]& \\ & \end{array}& 4)\end{array}$

[0059] Referring to FIG. 8, in one embodiment, the image processing circuit 50 of FIG. 4 filters the raw motion values RM to reduce the occurrence of false motion detection caused by noise and to limit the motion values to four bits, i.e., a maximum value of 15. FIG. 8 is a plot of the filtering algorithm according to an embodiment of the invention. Thus, the circuit 50 calculates the filtered motion values FM according to the following equation: $\begin{array}{cc}\{\begin{array}{cc}\mathrm{RM}\le 8& FM=0\\ 8<\mathrm{RM}<38& FM=\frac{\mathrm{RM}-8}{2}\\ \mathrm{RM}\ge 38& FM=15\end{array}& 5)\end{array}$

[0060] According to equation (5), the circuit 50 considers a raw motion value RM that is less than or equal to 8 to be noise, not true motion, and thus generates a corresponding filtered motion value FM=0. Likewise, the circuit 50 limits the maximum FM by clipping to 15 all RM values that are greater than or equal to 38.

[0061] Referring back to FIG. 4, because the image processing circuit 50 of FIG. 5 derives the luminance difference values DY from groups of four original pixels in the same original line, the circuit 50 cannot use the above-described technique to generate raw motion values for the filler pixels at the beginnings and ends of filler lines. For example, the circuit 50 uses four pixels P₀₁, P₀₂, P₀₃, and P₀₄ in the same line to generate some of the DY values for the filler block 80. Therefore, P₀₁ precedes the block 80 in a horizontal direction, and P₀₄ proceeds the block 80 in the horizontal direction. But referring to the filler blocks 82 and 84, because no pixels precede the block 82 and no pixels proceed the block 84, equation (1) is invalid for these blocks. Thus, in one embodiment, the circuit 50 calculates no raw or filtered motion values for the filler blocks 82 and 84 and the other filler blocks that contain the first two or last two filler pixels of respective filler lines. Alternatively, the circuit 50 assigns predetermined filtered motion values to these blocks. For example, in one embodiment, the circuit 50 assigns these blocks a constant FM value or the same FM value as calculated for the adjacent block in the same filler line. For example, according to the latter approach, the circuit 50 sets the filtered motion value FM₀₀ for the block 82 equal to FM₀₁, which the circuit 50 calculates for the adjacent block 80 as described above.

[0062] FIG. 9 illustrates the content layout of a motion-value buffer 90 for storing filtered motion values FM for filler fields derived from MPEG 4:2:0 original fields according to an embodiment of the invention. Referring to FIG. 4, in one embodiment, the image processing circuit 50 dedicates a portion of the memory 58 as the buffer 90, although the buffer 90 may reside in another memory. The circuit 50 includes only one buffer 90, and updates the contents of this buffer for each filler field. A procedure for updating the buffer 90 is discussed below in conjunction with FIG. 16.

[0063] The storage locations FM of the buffer 90 respectively correspond to the filler-pixel blocks described in conjunction with FIG. 7. For example, the location FM₀₁ stores the filtered motion value FM₀₁, which corresponds to the block 80 of FIG. 7. Furthermore, if the image processing circuit 50 of FIG. 4 assigns motion values to the beginning-line and ending-line pixel blocks, then the buffer 90 also includes optional locations that are shown in dashed line. For example, the optional location FM₀₀ corresponds to the beginning-line block 82 of FIG. 7, and the optional location FM_0(x/2) corresponds to the ending-line pixel block 84.

[0064] Referring to FIGS. 7 and 9, because the dimensions of the filler-pixel blocks such as the block 80 are 2×2 and because the image processing circuit 50 of FIG. 4 calculates one FM value per block, the horizontal dimension of the buffer 90 is either half or two pixels less than half the horizontal dimension of the original and filler fields. Specifically, if the motion-value buffer 90 includes the optional storage locations shown in dashed line, then the horizontal dimension of the buffer 90 is half the horizontal dimension of the original and filler fields. For example, if the original and filler fields have horizontal dimensions of x=720 pixels, then the buffer 90 is 720÷2=360 memory locations wide. Alternatively, if the motion-value buffer 90 does not include the optional storage locations shown in dashed line, then the horizontal dimension of the buffer 90 is half the horizontal dimension of the original and filler fields minus two pixels. For example, if the original and filler fields have horizontal dimensions of x=720 pixels, then the buffer 90 is (720÷2)−2=358 memory locations wide.

[0065] Similarly, the vertical dimension of the buffer 90 is one-half the vertical dimension of the original and filler fields, and thus one-fourth the vertical dimension of the resulting progressive frames generated by the image processing circuit 50 of FIG. 4. This is true whether or not the buffer 90 includes the optional storage locations. For example, if the original and filler fields each have vertical dimensions of k/2=240 lines—the corresponding progressive frames have k=2×240=480 lines—then the buffer 90 is k/4=240÷2—480÷4 with respect to the corresponding progressive frames—=120 memory locations high.

[0066] FIG. 10 illustrates the content layout of a motion-trace buffer 92 for storing motion-trace values MT for filler fields derived from MPEG 4:2:0 original fields according to an embodiment of the invention. As discussed below in conjunction with FIG. 16, each motion-trace value specifies the number of filler fields for which a respective FM value is valid. Referring to FIG. 4, in one embodiment, the image processing circuit 50 dedicates a portion of the memory 58 as the buffer 92, although the buffer 92 may reside in another memory. The circuit 50 includes only one buffer 92, and updates the contents of this buffer for each filler field. A procedure for updating the buffer 92 is discussed below in conjunction with FIG. 16.

[0067] The storage locations MT of the buffer 92 respectively correspond to the filler-pixel blocks described in conjunction with FIG. 7. For example, the location MT₀₁ stores the motion-trace value MT₀₁, which corresponds to the block 80 of FIG. 7, and thus which corresponds to the location FM₀₁ of the motion-value buffer 90 of FIG. 9. Furthermore, if the image processing circuit 50 of FIG. 4 assigns motion values, and thus motion-trace values, to the beginning-line and ending-line pixel blocks, then the buffer 92 also includes optional locations that are shown in dashed line. For example, the optional location MT₀₀ corresponds to the beginning-line block 82 of FIG. 7, and thus corresponds to the location FM₀₀ of the motion-value buffer 90. Similarly, the optional location MT_0(x/2) corresponds to the ending-line pixel block 84 of FIG. 7, and thus corresponds to the location FM_0(x/2) of the motion-value buffer 90.

[0068] Still referring to FIG. 10, the motion-trace buffer 92 has the same dimensions as the motion-value buffer 90 as discussed above in conjunction with FIG. 9. Furthermore, in one embodiment, each storage location MT is four bits wide.

[0069] Referring to FIGS. 9 and 10, the motion-value and motion-trace buffers 90 and 92 are significantly smaller than the motion memories of many prior image processing circuits. Furthermore, one can vary a motion-trace value within a predetermined range to vary the number of filler fields for which a respective FM value is valid without increasing the size of the buffer 92.

[0070] Referring to FIGS. 7, 9, and 10, as long as k+1 is divisible by four, then there are no partial (2×1) filler-pixel blocks at the bottom of the complimentary odd filler fields for the even fields E. For example, the last row of filler blocks include respective pixels from the filler lines a(k−2) and a(k), and there are no unpaired filler lines below this. Conversely, if k+1 is not divisible by four, then there is a row of partial filler blocks that include pixels from only one filler line a(k). In this situation, the image processing circuit 50 of FIG. 4 can calculate the raw and filtered motion values and the motion-trace values for these partial filler blocks in a number of ways. For example, the circuit 50 can set the raw and filtered motion values and the motion-trace value for a partial filler block equal to the raw and filtered motion values and motion-trace value, respectively, for the full filler block immediately above the partial filler block. For example, referring to FIGS. 7 and 9, if the filtered-motion-value location FM_(k,4)1 corresponds to a partial filler block in the filler field that compliments E₀, then the circuit 50 can set FM_(k/4)1=FM_((k/4)−1)1 and MT_(k/4)−1)1.

[0071] Referring to FIG. 11, the generation of motion values for filler pixels in odd filler fields is discussed according to an embodiment of the invention. For example purposes, the generation of motion values is discussed in conjunction with the original odd fields O₀ and O₁ being represented in a Y, C_B, and C_R color space and having been compressed according to the MPEG 4:2:0 format, it being understood that the same principles apply to the other original odd fields of the FIG. 6 sequence. The generation of motion values for filler pixels in even filler fields is discussed above in conjunction with FIG. 7.

[0072] FIG. 11 is a diagram of the original and filler pixels that compose the original and filler lines, respectively, of the odd fields O₀ and O₁. The pixels of the original odd field O₀ and its corresponding even filler field are denoted as P′_kx, and the pixels of O₁ and its corresponding even filler field are denoted as P′_kx, where k denotes the line and x denotes the column. Like the original pixels, the filler pixels are arranged in 2×2 blocks of four pixels (see FIG. 2A). For example, a block 94 includes filler pixels P₀₂, P₀₃, P₂₂, and P₂₃, which-compose the complimentary filler field for O₀.

[0073] Still referring to FIG. 11, the calculation of the difference values DY, DC_R, and DC_B and the raw and filtered motion values RM and FM for the even filler fields are similar to the respective DY, DC_R, DC_B, RM, and FM calculations for the odd filler fields as described above in conjunction with FIG. 7. For example, DY, DC_R, DC_B, and RM for the block 94 are given by the following equations: $\begin{array}{cc}{\mathrm{DY}}_{\mathrm{ij}}={Y_{\mathrm{ij}}-Y_{\mathrm{ij}}^{\prime }}_{i=1,3;j=1,2,3,4}& 6)\\ D{C}_{\mathrm{R01}}=C_{\mathrm{R01}}-C_{\mathrm{R01}}^{\prime }& 7)\\ D{C}_{\mathrm{B01}}=C_{\mathrm{B01}}-C_{\mathrm{B01}}^{\prime }& 8)\\ {\mathrm{RM}}_{01}=\mathrm{Max}\left[\frac{1}{8}\sum _{i_{\mathrm{odd}}=1}^2\sum _{j=0}^4{\mathrm{DY}}_{\mathrm{ij}},D{C}_{\mathrm{R01}},D{C}_{\mathrm{B01}}\right]& 9)\end{array}$
FM₀₁ is given by equation (5).

[0074] Referring to FIGS. 7, 9, and 11 and equations (5) and (9), the location FM₀₁ of the motion-value buffer 90 corresponds to the block 80 of FIG. 7 and to the block 94 of FIG. 11. Therefore, the image processing circuit 50 of FIG. 4 stores only one FM₀₁ value—FM₀₁ for the block 80 or FM₀₁ for the block 94—in the location FM₀₁. The procedure for selecting which FM₀₁ to store is discussed below in conjunction with FIG. 16.

[0075] Furthermore, as discussed above in conjunction with FIG. 7, the image processing circuit 50 of FIG. 4 cannot use the above-described technique to generate raw motion values for the filler blocks such as blocks 96 and 98 that include filler pixels at the beginnings and ends of filler lines. Thus, in one embodiment, the circuit 50 calculates no raw or filtered motion values for the filler blocks 96 and 84 and the other filler blocks containing the first two or last two filler pixels of respective filler lines. Alternatively, the circuit 50 assigns predetermined filtered motion values to these blocks. For example, in one embodiment, the circuit 50 assigns these blocks a constant FM value or the same FM value as calculated for the adjacent block in the same filler line. For example, according to the latter approach, the circuit 50 sets the filtered motion value FM₀₀ for the block 96 equal to FM₀₁, which the circuit 50 calculates for the adjacent block 94 as described above.

[0076] Referring to FIGS. 9, 10, and 11, as discussed above in conjunction with FIG. 7, as long as k+1 is divisible by four, then there are no partial (2×1) filler-pixel blocks at the bottom of the complimentary even filler fields for the odd fields O. Conversely, if k+1 is not divisible by four, then there is a row of partial filler blocks that include pixels from only one odd filler line b(k−1). In this situation, the image processing circuit 50 of FIG. 5 can calculate the raw and filtered motion values and the motion-trace values for these partial filler blocks in a number of ways as discussed above in conjunction with the filler field of FIG. 7.

[0077] Referring to FIG. 12, the generation of motion values for filler pixels in even filler fields is discussed according to another embodiment of the invention in which the original even fields of FIG. 6 are represented in a Y, C_B, and C_R color space and have been compressed and decompressed according to the MPEG 4:2:2 format. Like the original pixels, the filler pixels are arranged in 2×1 blocks of two pixels (see FIG. 3A). For example, a block 98 includes filler pixels P₁₂ and P₁₃, and a block 100 includes filler pixels P₃₂ and P₃₃. Thus, the major difference between FIGS. 7 and 12 is that in FIG. 12, the filler-pixel blocks contain two pixels instead of four pixels. Therefore, DY, DC_R, DC_B, and RM for the block 98 are given by the following equations:
DY_ij=|Y_ij−Y′_ij|_{i=0,2;j=1,2,3,4}  10)
DC_Ri1=C_Ri1−C′_Ri1|_i=0,2  11)
DC_Bi1=C_Bi1−C′_Bi1|_i=0,2  12)
$\begin{array}{cc}{\mathrm{RM}}_{01}=\mathrm{Max}\left[\frac{1}{8}\sum _{i_{\mathrm{even}}=0}^2\sum _{j=1}^4{\mathrm{DY}}_{\mathrm{ij}},\frac{1}{2}\sum _{i_{\mathrm{even}}=0}^{2}D{C}_{R_{i}1},\frac{1}{2}\sum _{i_{\mathrm{even}}=0}^{2}D{C}_{B_{i}1}\right]& 13)\end{array}$
FM₀₁ is given by equation (5). Furthermore, the calculation of DY is the same as for the 4:2:0 format, and thus equation (10) is the same as equation (5). Furthermore, because the 4:2:2 format calls for one C_R and one C_B value for each 2×1 block of original pixels, the calculation of DC_R includes taking the difference between C_R01, which corresponds to pixels P₀₂ and P₀₃ of E₀, and C′_R01, which corresponds to P′₀₂ and P′₀₃ of E₁, and taking the difference between C_R11, which corresponds to pixels P₂₂ and P₂₃ of E₀, and C′_R11, which corresponds to pixels P′₂₂ and P′₂₃ of E₁. A similar analysis applies to DC_B.

[0078] Still referring to FIG. 12, the image processor circuit 50 of FIG. 5 calculates difference and raw motion values for the 2×1 block 100 according to the following equations:
DY_ij=|Y_ij−Y′_ij|_{i=2,4; j=1,2,3,4}  14)
DC_Ri1=|C_Ri1−C′_Ri1|_i=2,4  15)
DC_Bi1=|C_Bi1−C′_Bi1|_i=2,4  16)
$\begin{array}{cc}{\mathrm{RM}}_{11}=\mathrm{Max}\left[\frac{1}{8}\sum _{i_{\mathrm{even}}=2}^4\sum _{j=1}^4{\mathrm{DY}}_{\mathrm{ij}},\frac{1}{2}\sum _{i_{\mathrm{even}}=2}^{4}D{C}_{R_{i}1},\frac{1}{2}\sum _{i_{\mathrm{even}}=2}^{4}D{C}_{B_{i}1}\right]& 17)\end{array}$

[0079] Furthermore, for the same reasons discussed above in conjunction with FIG. 7, the image processing circuit 50 of FIG. 4 cannot use the above-described technique to generate raw or filtered motion values for the filler-pixel blocks 102 and 104 and other filler blocks containing filler pixels at the beginnings and ends of filler lines. Thus, the circuit 50 generates filtered motion values for these filler blocks as discussed above in conjunction with FIG. 7.

[0080] In addition, if k+1 is divisible by two, the last filler line a(k) of the filler field that compliments E₀ is not “sandwiched” between two original lines of E₀. Therefore, the circuit 50 calculates DY, DC_R, DC_B, and RM for a last-line filler block such as the block 106 using original pixels in only the last lines a(k−1) and c(k−1), respectively, of the original fields E₀ and E₁. For example, the circuit 50 calculates the difference and raw motion values for the pixel block 106 according to the following equations:
DY_(k−1)j=|Y_(k−1)j−Y′_(k−1)j|_j=1,2,3,4  18)
DC_R(k−1)1=C_R(k−1)1−C′_R(k−1)1|  19)
DC_B(k−1)1=C_B(k−1)1−C′_B(k−1)1|  20)
$\begin{array}{cc}{\mathrm{RM}}_{\left(k-1\right)1}=\mathrm{Max}\left[\frac{1}{4}\sum _{j=1}^4{\mathrm{DY}}_{\left(k-1\right)j},D{C}_{R\left(k-1\right)1},D{C}_{B\left(k-1\right)1}\right]& 21)\end{array}$
Thus, for example DY for the block 106 is calculated using the luminance values for the pixels P_(k−1)1, P_(k−1)2, P_(k−1)3, and P_(k−1)4 from E₀ and P′_(k−1)0, P′_(k−1)1, P′_(k−1)2, P′_(k−1)3, and P′_(k−1)4 from E₁. Likewise, DC_R and DC_B are calculated from the C_R and C_B values, respectively, for the 2×1 blocks of original pixels that include P_(k−1)2 and P_(k−1)3 from E₀ and P′_(k−1)2 and P′_(k−1)3 from E₁. The circuit 50 uses equation (5) to calculate the filtered motion values.

[0081] FIG. 13 illustrates the content layout of a motion-value buffer 108 for storing filtered motion values FM for filler fields derived from MPEG 4:2:2 original fields according to an embodiment of the invention. Referring to FIG. 4, in one embodiment, the image processing circuit 50 dedicates a portion of the memory 58 as the buffer 108, although the buffer 108 may reside in another memory. The circuit 50 includes only one buffer 108, and updates the contents of this buffer for each filler field. A procedure for updating the buffer 108 is discussed below in conjunction with FIG. 16.

[0082] Referring to FIGS. 12 and 13, because the dimensions of the filler-pixel blocks such as the block 98 are 2×1 and because the image processing circuit 50 of FIG. 4 calculates one FM value per block, the horizontal dimension of the buffer 108 is either half or two pixels less than half the horizontal dimension of the original and filler fields. Specifically, if the motion-value buffer 108 includes the optional storage locations shown in dashed line, then the horizontal dimension of the buffer 108 is half the horizontal dimension of the original and filler fields. For example, if the original and filler fields have horizontal dimensions of x=720 pixels, then the buffer 108 is X/2=720÷2=360 memory locations wide. Alternatively, if the motion-value buffer 108 does not include the optional storage locations shown in dashed line, then the horizontal dimension of the buffer 108 is half the horizontal dimension of the original and filler fields minus two pixels. For example, if the original and filler fields have horizontal dimensions of x=720 pixels, then the buffer 90 is (x/2)−2=(720÷2)−2=358 memory locations wide.

[0083] Similarly, the vertical dimension of the buffer 108 is the same as the vertical dimension of the original and filler fields, and thus one-half the vertical dimension k of the resulting progressive frames generated by the image processing circuit 50 of FIG. 4. This is true whether or not the buffer 108 includes the optional storage locations. For example, if the original and filler fields each have vertical dimensions of k/2=240 lines—the corresponding progressive frames have k=2×240=480 lines—then the buffer 108 is k/2=240−480+2 with respect to the corresponding progressive frames—=240 memory locations high.

[0084] Thus, the motion-value buffer 108 has the same horizontal dimension and twice the vertical dimension as the motion-value buffer 90 of FIG. 9.

[0085] FIG. 14 illustrates the content layout of a motion-trace buffer 110, which is similar to the buffer 92 of FIG. 10 except that it stores motion-trace values MT for filler fields derived from MPEG 4:2:2 original fields according to an embodiment of the invention. The storage locations MT of the buffer 110 respectively correspond to the filler-pixel blocks described in conjunction with FIG. 12. For example, the location MT₀₁ stores the motion-trace value MT₀₁, which corresponds to the block 98 of FIG. 12, and thus which corresponds to the location FM₀₁ of the motion-value buffer 108 of FIG. 13. Furthermore, if the image processing circuit 50 of FIG. 4 assigns motion values, and thus motion-trace values, to the beginning-line and ending-line pixel blocks, then the buffer 108 also includes optional locations that are shown in dashed line. For example, the optional location MT₀₀ corresponds to the beginning-line block 102 of FIG. 12, and thus corresponds to the location FM₀₀ of the motion-value buffer 108.

[0086] Still referring to FIG. 14, the motion-trace buffer 100 has the same dimensions as the motion-value buffer 108 as discussed above in conjunction with FIG. 13.

[0087] Referring to FIG. 15, the generation of motion values for filler pixels in odd filler fields is discussed for the original odd fields of FIG. 6 being represented in a Y, C_B, and C_R color space and having been compressed and decompressed according to the MPEG 4:2:2 format. The calculation of the difference values DY, DC_R, and DC_B and the raw and filtered motion values RM and FM for the even filler fields are similar to the respective DY, DC_R, DC_B, RM, and FM calculations for the even filler fields as described above in conjunction with FIG. 12. For example, DY, DC_R, DC_B, and RM for the block 112 are given by the following equations:
DY_ij=|Y_ij−Y′_ij|_{i=1,3;j=1,2,3,4}  22)
DC_Ri1=|C_Ri1−C′_Ri1|_i=1,3  23)
DC_Bi1=|C_Bi1−C′_Ri1|_i=1,3  24)
$\begin{array}{cc}{\mathrm{RM}}_{11}=\mathrm{Max}\left[\frac{1}{8}\sum _{i_{\mathrm{odd}}=1}^3\sum _{j=0}^4{\mathrm{DY}}_{\mathrm{ij}},\frac{1}{2}\sum _{i_{\mathrm{odd}}=1}^{3}D{C}_{R_{i}1},\frac{1}{2}\sum _{i_{\mathrm{odd}}=1}^{3}D{C}_{B_{i}1}\right]& 25)\end{array}$

[0088] FM₀₁ is given by equation (5). Furthermore, for the same reasons discussed above in conjunction with FIG. 7, the image processing circuit 50 of FIG. 4 cannot use the above-described technique to generate raw or filtered motion values for the filler-pixel blocks 114 and 116 and other filler blocks containing filler pixels at the beginnings and ends of filler lines. Thus, the circuit 50 generates filtered motion values for these filler blocks as discussed above in conjunction with FIG. 7.

[0089] In addition, because the first filler line b0 of the filler field that compliments O₀ is not “sandwiched” between two original lines of O₀, the circuit 50 calculates DY, DC_R, DC_B, and RM for a first-line filler block such as the block 118 using original pixels in only the second lines b1 and d1, respectively, of the original fields O₀ and O₁. For example, the circuit 50 calculates the difference and raw motion values for the pixel block 118 according to the following equations:
DY_1j=|Y_1j−Y′_1j|_j=1,2,3,4  26)
DC_R01=|C_R01−C′_R01|  27)
DC_B01=|C_B01−C′_B01|  28)
$\begin{array}{cc}{\mathrm{RM}}_{01}=\mathrm{Max}\left[\frac{1}{4}\sum _{j=1}^4{\mathrm{DY}}_{1j},D{C}_{\mathrm{R01}},D{C}_{\mathrm{B01}}\right]& 29)\end{array}$

[0090] FM₀₁ is given by equation (5). Similarly, if k+1 is not divisible by two, the last filler line bk of the filler field is not “sandwiched” between two original lines of O₀. Therefore, the circuit 50 calculates DY, DC_R, DC_B, and RM for a last-line filler block using original pixels in only the lines b(k−1) and d(k−1), respectively.

[0091] Referring to FIGS. 12, 13, and 15, the location FM₀₁ of the motion-value buffer 108 corresponds to the block 98 of FIG. 12 and to the block 108 of FIG. 15. Therefore, the image processing circuit 50 of FIG. 4 stores only one FM₀₁ value—FM₀₁ for the block 98 or FM₀₁ for the block 108—in the location FM₀₁. The procedure for selecting which FM₀₁ to store is discussed below in conjunction with FIG. 16.

[0092] FIG. 16 is a flow diagram of the technique that the image processing circuit 50 of FIG. 4 implements to initialize and update the motion-value buffer 90 and motion-trace buffer 92 of FIGS. 9 and 10 (4:2:0 format) or the buffers 108 and 100 of FIGS. 13 and 14 (4:2:2 format) according to an embodiment of the invention. For clarity, this technique is discussed in conjunction with the buffers 90 and 92, it being understood that the technique is similar for the buffers 108 and 110.

[0093] Referring to FIG. 4 and block 112 of FIG. 16, the processor 56 loads 0 into all storage locations of the buffers 90 and 92.

[0094] Referring to block 114, the processor 56 then retrieves the next filtered motion value FM_kx, which it is previously calculated and stored in the memory 58.

[0095] Referring to blocks 116 and 118, if FM_kx is greater than or equal to the current contents of the location k, x of the motion-value buffer 90, then the processor 56 overwrites the location k, x with FM_kx. Next, referring to block 120, the processor 56 loads an initial MT value into the k, x location of the motion-trace buffer 92. In one embodiment, the initial MT value equals 5.

[0096] Conversely, referring to blocks 116 and 122, if FM_kx is less than the current contents of the k, x location of the motion-value buffer 90, then the processor 56 analyzes the contents of the k, x, location of the trace buffer 92. If the contents equals 0, then, referring to block 124, the processor 56 loads 0 into the k, x location of the motion-value buffer 90 to indicate that there is no motion associated with the respective filler-pixel block of the current filler field. Conversely, referring to block 126, if the contents of the k, x location of the trace buffer 92 does not equal 0, then the processor 56 decrements the contents by a predetermined value D. In one embodiment, D=1.

[0097] Referring to block 128, the processor 56 processes the next FM_kx value in a similar manner.

[0098] Therefore, by varying the values of D and the initial MT, one can vary the maximum number of filler fields that a motion value will influence. For example, referring to FIGS. 6, 7, 9, 10, 11, and 16, suppose that the initial MT=5, D=1, FM₀₁=15 for the filler-pixel block 80 of the filler line a1 for E₀, and FM₀₁<15 for the corresponding filler-pixel blocks of the filler lines b0-b2, c1-c3, d0-d2, e1-e3, and f0-f2 for O₀, E₁, O₁, E₂, and O₂, respectively. Thus, according to the flow diagram of FIG. 16, for the block 80, the processor 56 loads FM₀₁=15 into the FM₀₁ location of the motion-value buffer 90 and loads 5 into the MT₀₁ location of the trace buffer 92. Next, because FM₀₁<15 for the filler-pixel block 94 (FIG. 11), the processor 56 leaves the previous FM₀₁=15 in the FM₀₁ location of the buffer 90 and decrements the contents of the MT₀₁ location of the buffer 92 to 4. The processor 56 processes FM₀₁<15 for the filler-pixel blocks (not shown) for c1-c3, d0-d2, e1-e3, and f0-f2 in a similar manner. After processing the filler-pixel block of f0-f2, however, the location MT₀₁ of the trace buffer 92 equals 0. Thus, as discussed below, even if the processor 56 detects no subsequent motion, the motion detected between E₀ and E₁ influences the values of the filler pixels in the filler fields complimentary to six consecutive original fields: E₀, O₀, E₁, O₁, E₂, and O₂. Thus, unlike many of the prior image processing circuits, the image processing circuit 50 allows detected motion to influence the filler-pixel values in more than four filler fields. Furthermore, one can vary the initial MT value or D to vary the number of motion-affected filler fields without increasing the sizes the buffers 90 and 92.

[0099] FIG. 17 is a diagram of the original pixels that the image processing circuit 50 of FIG. 4 uses to calculate direction values and to spatially interpolate a pixel value for a filler pixel according to an embodiment of the invention. For example purposes, the filler pixel is P₃₃ of the filler line a3 of FIG. 7, it being understood that the following discussion applies to other filler pixels except the first and last pixels of each filler line. Calculating values for these filler pixels is discussed below. As discussed below in conjunction with FIGS. 18A-18E, the circuit 50 calculates the direction values and spatially interpolates the pixel value for P₃₃ from the three original pixels P₂₂, P₂₃, and P₂₄ above P₃₃ and three pixels P₄₂, P₄₃, and P₄₄ below P₃₃.

[0100] Referring to FIGS. 18A-18E, in one embodiment of the invention, the image processing circuit 50 recognizes three edge directions and two thin-line directions with respect to the pixel diagram of FIG. 17. In determining the direction values, the circuit 50 uses only the luminance values Y of the original pixels in the pixel diagram.

[0101] Referring to FIG. 18A, the circuit 50 recognizes a 45°-225° edge if the original pixels P₂₄ and P₄₂ have similar Y values. For example, if the normal to the edge points toward the lower right of the pixel group, then the pixels P₂₂ and P₂₃ have Y values similar to the Y values of P₄₂ and P₂₄. Conversely, if the normal to the edge points toward the upper left of the pixel group, then the pixels P₄₃ and P₄₄ have Y values similar to the Y values of P₂₄ and P₄₂.

[0102] Referring to FIG. 18B, the circuit 50 recognizes a thin line that extends through the pixels P₂₃, P₂₄, P₄₂, and P₄₃ at approximately 63°-243°. This thin line is characterized by P₂₂ and P₄₄ having Y values that are significantly different than the Y values of P₂₃, P₂₄, P₄₂, and P₄₃. Referring to FIG. 18C, the circuit 50 recognizes a 90°-270°edge if the original pixels P₂₃ and P₄₃ have similar Y values. For example, if the normal to the edge points toward the right of the pixel group, then the pixels P₂₂ and P₄₂ have Y values similar to the Y values of P₂₃ and P₄₃. Conversely, if the normal to the edge points toward the left of the pixel group, then the pixels P₂₄ and P₄₄ have Y values similar to the Y values of P₂₃ and P₄₃. The circuit 50 also recognizes a 90°-270° edge if all of the pixels P₂₂, P₂₃, P₂₄, P₄₂, P₄₃, and P₄₄ have similar Y values.

[0103] Referring to FIG. 18D, the circuit 50 recognizes a thin line that extends through the pixels P₂₂, P₂₃, P₄₃, and P44 at approximately 117°-297°. This thin line is characterized by P₂₄ and P₄₂ having Y values that are significantly different than the Y values of P₂₂, P23, P₄₃, and P₄₄.

[0104] Referring to FIG. 18E, the circuit 50 recognizes a 135°-315° edge if the original pixels P₂₂ and P₄₄ have similar Y values. For example, if the normal to the edge points toward the lower left of the pixel group, then the pixels P₂₃ and P₂₄ have Y values similar to the Y values of P₂₂ and P₄₄. Conversely, if the normal to the edge points toward the upper right of the pixel group, then the pixels P₄₂ and P₄₃ have Y values similar to the Y values of P₂₂ and P₄₄.

[0105] The direction values DV are calculated according to column 2 of Table I.

[0106]

TABLE I
		If minimum of pixel
		differences is below
		Tedge threshold,
		estimate missing
Directions	Pixel difference along directions	pixel value Ps as:
45°-225°	DV45-225 = |P24 − P42| + offset	Ps33 = (P24 + P42)/2
(FIG.
18A)
63°-243°	DV63-243 = (|P24 − P43| + |P23 −	Ps33 = (P24 + P42 +
(FIG.	P42|)/2
18B)	P23 + P43)/4
90°-270°	DV90-270 = |P23 − P43|	Ps33 = (P23 + P43)/2
(FIG.
18C)
117°-297°	DV117-297 = (|P23 − P44| + |P22 −	Ps3 = (P22 + P44 +
(FIG.	P43|)/2
18D)	P23 + P43)/4
135°-315°	DV135-315 = |P22 − P44|+ offset	Ps = (P22 + P44)/2
(FIG.
18E)
No dominant edge direction if min. of above	Minimum DV >
values >Tedge	Tedge so
	estimate Ps as:
	Ps = (P23 + P43)/2

[0107] The circuit 50 calculates all the DV values according to column 2 of Table I, identifies the minimum DV value, and compares the minimum DV to a threshold T_edge. If the minimum DV is greater than T_edge, then processor 56 identifies by default a 90°-270° edge (FIG. 18C), and thus spatially interpolates Y, C_R, and C_B values (collectively represented as P_s in Table I) for P₃₃ equal to the average of the respective Y, C_R, and C_B values of P₂₃ and P₄₃ as shown in the last row of column 3 of Table I. Conversely, if the minimum DV is less than or equal to T_edge, then the processor 56 calculates P_s according to the equation corresponding to the minimum DV. For example, if DV_45-225 is the minimum, then the processor 56 calculates P_s equal to the average of the values of P₂₄ and P₄₂. If DV_63-243 is the minimum value, then the processor 56 calculates P_s equal to the average of the values of P₂₃, P₂₄, P₄₂, and P₄₃. If DV_90-270 is the minimum value, then the processor 56 calculates P_s equal to the average of the values of P₂₃ and P₄₃. If DV_117-297 is the minimum value, then the processor 56 calculates P_s equal to the average of the values of P₂₂, P₂₃, P₄₃, and P₄₄. And if DV_135-315 is the minimum value, then the processor 56 calculates P_s equal to the average of the values of P₂₂ and P₄₄.

[0108] T_edge is an empirically determined constant. In one embodiment, it is in the range of approximately 40-50.

[0109] The inventor has determined that the 63°-243° and 117°-297° thin lines have a tendency to be misinterpolated as 45°-225° and 135°-315° edges. Therefore, the circuit 50 adds an offset to DV_135-315 and DV_45-225 to effectively offset this misinterpolation by favoring the detection of thin lines. In one embodiment, the processor 56 calculates the offset according to the following equation: $\begin{array}{cc}\mathrm{offset}=\{\begin{array}{cc}10~20& \mathrm{when}P_{22}-P_{24}>t_{\mathrm{line}}\\ & \mathrm{and}P_{42}-P_{44}>t_{\mathrm{line}}\\ 0& \mathrm{Otherwise}\end{array}& 30)\end{array}$
where t_line is a threshold empirically determined to be approximately 30 in one embodiment of the invention.

[0110] Referring to FIGS. 7, 11, and 17, the processor 56 also temporally interpolates Y, C_R, and C_B (collectively P_t) for each filler pixel from the corresponding original pixels in an adjacent original field having the same polarity as the filler field. For example, the processor 56 calculates P_t for the filler pixel P₃₃ in the filler line a3 of E₀ (FIG. 7) equal to the luminance and chromanance values of the original pixel P₃₃ in the original line b3 of O₀ (FIG. 11).

[0111] Next, the processor 56 calculates the final values P_f of the filler pixels according to the following equations:
α=max(FM_(i−1)j,FM_ij,FM_(i+1)j)  31)
$\begin{array}{cc}{P}_f=\frac{1}{15}\left(\alpha P_s+\left(15-\alpha \right)P_t\right)& 32)\end{array}$

[0112] Specifically, the processor 56 calculates P_f equal to the sum of α-weighted P_s and (1−α) weighted P_f. α equals the maximum of the FM value of the filler pixel for which P_f is being calculated, and the FM values of the filler-pixel blocks above and below the filler pixel. For example, referring to FIGS. 7 and 9, α for the filler pixel P₃₃ is the maximum of FM₀₁ and FM₁₁ (there is no filler pixel block above the block 80, and thus no FM location of the buffer 90 above FM₀₁) Taking the maximum of the three closest FM values in a vertical direction ensures that the spatially interpolated value P_s is given the greatest weight where there is detected motion. Furthermore, as stated above in conjunction with FIG. 8, the maximum value of FM, and thus the maximum value of α, is 15. Therefore, the right side of equation (32) is divided by 15 for normalization.

[0113] Referring to equation (32), if there is significant motion such that α is relatively large, then P_f is weighted heavily toward the spatially interpolated value P_s. Conversely, if there is little motion such that α is relatively small, then P_f is weighted heavily toward the temporally interpolated value P_t.

[0114] Referring to FIG. 7, the generation of values for filler pixels at the beginnings and ends of filler pixel lines is discussed. For example purposes, the filler pixel P₁₀ of E₀ is discussed, it being understood that the following discussion applies to P₁₁ and other filler pixels at the beginning and ending of filler lines.

[0115] If the image processing circuit 50 assigns a predetermined value FM₀₀ to P₁₀ a and the other beginning- and end-of-line filler pixels as described above, then the processor 56 calculates P_s, P_t, and P_f for P₁₀ as discussed above.

[0116] Conversely, if the processing circuit 50 does not assign a predetermined value FM₀₀ to P₁₀ and the other beginning- and end-of-line filler pixels as described above, then the processor 56 calculates P_f a manner other than that described above. For example, the processor 56 may exclusively spatially interpolate P_f for P₁₀ equal to the average of the respective Y, C_R, and C_B values of the vertically adjacent original pixels P₀₀ and P₂₀. Or, the processor 56 may exclusively temporally interpolate P_f for P₁₀ equal to the Y, C_R, and C_B values of the corresponding original pixel P₁₀ in the adjacent original odd field O₀.

[0117] From the foregoing it will be appreciated that, although specific embodiments of the invention have been described herein for purposes of illustration, various modifications may be made without deviating from the spirit and scope of the invention.