Weighted absolute difference based deinterlace method and apparatus

[0001] This application claims priority from and is a continuation of U.S. provisional patent application Ser. No. 60/366,721, filed Mar. 21, 2002. We incorporate all priority applications by reference.

[0002] 1. Field of the Invention

[0003] The invention relates to a weighted absolute difference based deinterlace method and apparatus.

[0004] 2. Description of the Related Art

[0005] The National Television Standards Committee (NTSC) is responsible for setting television video standards in the United States. The NTSC standard is interlaced. Each frame (480 viewable lines) is broken into two sections called fields (240 alternating viewable lines). A field containing even lines (0, 2, 4, . . . ) is termed an even field. Similarly, a field containing odd lines (1, 3, 5, . . . ) is termed an odd field.

[0006] Every 1/60 second, the NTSC standard requires displaying an even or an odd field. Because this process is happening quickly, the eye is tricked into seeing a full frame picture. The image quality resulting from the NTSC standard is acceptable on smaller television sets—those having roughly 27 inches viewable area. The image quality, however, quickly degrades as the screen size increases. This is because the NTSC standard creates artifacts, e.g., flickering, shimmering, and diagonal line jaggedness, visible to the eye on large screens. This image quality degradation renders the NTSC standard also unacceptable for high definition television (HDTV). And NTSC is incompatible with most personal computer and liquid crystal (LC) displays that operate on progressive scan. Progressive scan presents a higher quality alternative to the interlaced NTSC standard. Progressive scan require displaying alternating odd and even lines with each frame.

[0007] Accordingly, a need remains for an improved method and apparatus that converts interlaced to progressive scan signals.

[0020] Converting interlaced video signals into progressively scanned video signals requires converting interlaced scanned fields, each comprising odd or even scan lines of a video signal frame, into progressively scanned fields, each comprising both odd and even scan lines.

[0021] The invention relates to a deinterlace method and apparatus adapted to interpolate pixel data for a missing line in an interlaced video signal. More particularly, the invention relates to a deinterlace method and apparatus using weighted absolute differences along different directions as means for interpolating pixel data using edge orientation detection.

[0022] FIG. 1 is a block diagram of a deinterlace apparatus. Referring to FIG. 1, the deinterlace apparatus comprises a line memory 10 adapted to store a predetermined number of lines of an input luminance signal Yin's future field. Each line of the future field includes a multitude of pixels. The future field has a first polarity, e.g., even.

[0023] The input luminance signal Yin passes to a first or current field memory 20 through 11. The current field memory 20 stores a current field of the input luminance signal Yin. The current field has a second polarity, e.g., odd.

[0024] The input luminance signal Yin passes to a second or previous field memory 30 through 21. The previous field memory 30 stores a previous field of the input luminance signal Yin. The previous field has a first polarity, e.g., even, same as the future field stored in the line memory 10.

[0025] The line memory 10 and the field memories 20 and 30 might be synchronous dynamic random access memory (SDRAM) or other types of memory. A person of reasonable skill in the art should recognize that the line memory 10 and the field memories 20 and 30 might be implemented in a singular memory, in distinct or separate memory, or any combination thereof.

[0026] Referring to FIG. 2A, the line memory 10 includes a future field sub image 40 having a predetermined size, e.g., 3×5. The future field sub image 40 is extracted from the line memory 10 through line 13. The future field sub image 40 is centered at pixel R34, located in the same position as the pixel to be interpolated (x in FIG. 3) in the current field. For convenience only, FIG. 2A shows a larger 3×9 future field sub image 40.

[0027] Referring to FIG. 2B, the field memory 30 includes a similar previous field sub image 60 having a predetermined size, e.g., 3×5. The previous field sub image 60 is extracted from the field memory 30 through line 31. The previous field sub image 60 is centered at pixel Q34, located in the same position as pixel R34 in the future sub image 40. Again for convenience only, FIG. 2B shows a larger 3×9 previous field sub image 60.

[0028] Referring to FIG. 3, the field memory 20 includes a current field sub image 50 having a predetermined size, e.g., 4×9. The current field sub image 50 is extracted from the field memory 20 through line 23. The current field sub image 50 is centered at an interpolated pixel represented by x. The sub image 50 includes, for example, even lines 0, 2, 4, and 6. Since the current field has a different polarity than the future or previous fields, the lines in the future or previous sub images 40 and 60, respectively, are odd lines 1, 3, and 5.

[0029] A person of reasonable skill in the art should recognize that any size of sub image comes within the scope of the present invention.

[0030] Referring to FIGS. 1, 2A, 2B, and 3, the pixel to be interpolated, denoted by x, is at the center of the sub images 40, 50, and 60. The pixel x is located on a missing line of the current field 50. When the pixel x is on a line that is close to a boundary of the video image, some pixels in any sub image might fall outside the video image. For example, suppose the current field 50 is an odd field that contains pixels of odd lines 1, 3, 5, . . . . And suppose that the pixel x is the top left corner of the video image, e.g., P00. This locates the center of the sub images at the top left corner of the video image. All of the pixels above and the left of P00 will be outside the video image. In this circumstance, the deinterlace method and apparatus sets the value of all pixels outside the video image to a predetermined value, e.g., zero.

[0031] The motion detector 200 detects motion between the future and previous field sub images 40 and 60. The motion detector 200 includes a subtractor 202 and an absolute operator 204. The subtractor 202 subtracts the luminance or value of a pixel in the future field sub image 42, e.g., pixel R34, from the luminance or value of a pixel in the previous field sub image 62, e.g., pixel Q34, where the pixels are located in the same position in the two different fields. That is, the subtractor 202 subtracts the future field sub image 42 from the previous field sub image 62 on a pixel-by-pixel basis through lines 41 and 63. The operator 204 takes the absolute of the difference produced by the subtractor 202 through line 203. The operator 204 provides the absolute difference to first and second comparators 210 and 220 through lines 207 and 205, respectively. The first comparator 210 compares the absolute difference at 207 to a first predetermined threshold 214. Likewise, the second comparator 220 compares the absolute difference at 205 to a second predetermined threshold 224. The first threshold 214 is typically set higher than the second threshold 224.

[0032] The first comparator 210 provides its results to a first counter 230 through line 211. The first counter 230 counts the number of cases where |Rij−Qij|>T214 in the sub image 42, where T214 is the first threshold 214. The first counter 230 provides its results to a third comparator 250. The third comparator asserts a motion signal at 251—indicating motion between future and previous sub images 40 and 60, respectively—when the results of the first counter 230 exceed a third threshold 254.

[0033] The second comparator 220 provides its results to a second counter 240 through line 221. The second counter 240 counts the number of cases where |Rij−Qij|<T224 in the sub image 42, T224 is the second threshold 224. The second counter 240 provides its results to a fourth comparator 260 through line 241. The fourth comparator 260 asserts a no motion signal at 263—indicating that the future and previous sub images 40 and 60, respectively, are part of a still (or non moving) portion of the video image—when the results of the second counter 240 exceed a fourth threshold 264.

[0034] The motion detector 200 includes a film mode detector 300 adapted to detect a film mode in the input luminance signal Yin. The film mode detector 300 comprises a film mode counter 270, comparator 280, and threshold 284. The film mode counter 270 counts the number of cases where the pixel to be interpolated x does not belong to a moving part of the image in a predetermined portion of the video image. The predetermined portion of the video image might be the whole field, the most active part of the video, an upper portion of the video, and the like. A person of reasonable skill in the art should understand that other criteria for selecting predetermined portions of the video image come within the scope of the present invention.

[0035] The film mode comparator 280 receives the counter 270's count at 271. The comparator 280 compares the count at 271 with the film mode threshold 284. The comparator 280 asserts a film mode signal at 281 if the count is less than the threshold 284. That is, the previous and the future fields come from the same frame indicating that the video was converted from film using the well known 3:2 pull down conversion method.

[0036] The first, second, third, fourth, and film mode thresholds 214, 224, 254, 264, 284, respectively, might be set through firmware, hardware, software, or a combination thereof.

[0037] The switch 100 selects output pixels at 71, 95, 81, 43, or 63 responsive to the motion signal at 251, the no motion signal at 263, and the film mode signal at 281. The switch 100 selects the spatially interpolated pixels at 71 responsive to the motion signal at 251. In this case, the correlation of the current field to the previous or future field is low. The effect of the previous or future field is, therefore, minimized. A purely spatial directional deinterlacing from the weighted absolute difference based spatial interpolator 70 is applied to interpolate the pixel of interest through 71.

[0038] The switch 100 selects the averaged signals at 95 responsive to the no motion signal at 263. The pixels at 95 are the average of R34 and Q34. Pixels from the future and previous sub image fields 42 and 62, respectively, are provided through line 45 and 61 to an adder 90.

[0039] The results of adder 90 are provided to a shift circuit 94 that outputs the average to line 95.

[0040] Put differently, a static mesh is applied to interpolate the pixel of interest x to increase the vertical resolution. Taking the average pixel value from the previous and future field sub images 40 and 60 allows about a 3 dB improvement in noise elimination.

[0041] The switch 100 selects spatial-temporally interpolated pixels at 81 responsive to the unasserted (or false) motion and no motion signals at 251 and 263, respectively. That is, if both the motion and no motion signals at 251 and 263 are false, the confidence that the pixel x to be interpolated belongs to either a moving or a still portion of the image is not high enough. The deinterlace apparatus, therefore, adopts a hybrid interpolation approach selecting spatial-temporally interpolated pixels at 81.

[0042] The switch 100 selects pixels at either 43 or 63 responsive to the film mode signal at 281, as we explain below. An asserted (or true) film mode signal at 281 will block the motion and no motion signals at 251 and 263, respectively, and thus, prevent the switch 100 from selecting the temporally interpolated, averaged, and spatial-temporally interpolated pixels at 71, 95, and 81, respectively. And an asserted film mode signal at 281 will disable the film mode detector 300 for four fields. For the next four fields, the deinterlace apparatus of FIG. 1 enters a film mode. During the film mode, the switch 100 selects either pixel R34 or pixel Q34 as the output pixel, depending on the inverse 3:2 pull down requirement. That is, the deinterlace apparatus will merge the current field with either the previous or future fields (since the previous and future fields are theoretically identical).

[0043] FIG. 4 is a field merge pattern. Referring to FIG. 4, the upper circles represent the input fields. The letter above them represents the polarity and the frame number from the film source. For example, field E0 if from frame 0 of the film source and its polarity is even.

[0044] The lower circles represent the output frames. The number beside each circle represents the output frame count. The letters B or F below each lower circle represent the merge pattern. The letter B represents a backward merge that merges the current field with the previous field. The letter F represents a forward merge that mergers the current field with the future field. The two arrows toward each output frame represent that the frame comes from those two fields indicated by the two arrows. The input field pattern shown in FIG. 4 complies with a typical 3:2 pull down format. The video generates three fields from frame 0 of the source film then two fields from frame 1, then three fields from frame 2, then two fields from frame 3, and so on.

[0045] At the position of the output frame 0, the current, previous, and future fields are from frame 0 of the source film. The previous and future fields are theoretically identical. In practice, however, noise prevents the previous and future fields from being identical. The deinterlace apparatus detects identical fields by scanning only a portion (a sub image) of the field. In other words, in the deinterlace apparatus, it is unnecessary to scan the field in its entirety to determine whether the video originated from film. When the deinterlace apparatus detects identical fields, it immediately enters film mode. Before the apparatus completes its film mode detection, therefore, it calculates the missing lines of frame 0. Once in film mode, the apparatus fills the remaining missing lines of frame 0 by merging the corresponding lines in the future field. The result is a partially merged frame 0 that we represent in FIG. 4 with a dotted line.

[0046] The film mode detector 300 is disabled for the next four fields, corresponding to output frames 1, 2, 3, and 4. The deinterlace apparatus merges the current field with the previous field for output frames 1 and 3 and merges the current field with the future field for output frames 2 and 4. At the fifth field, which corresponds to output frame 5, the film mode detector is enabled, and the film mode signal remains asserted (true) until the count at 271 provided by counter 270 exceeds the threshold 284 again. If the count at 271 does not exceed the film mode threshold 284 for a predetermined region, the film mode detector 300 again detects identical future and previous fields, and the deinterlace apparatus again enters film mode for the next four fields. If the film mode detector 200 does not detect identical future and previous fields, the count at 271 soon—within just a few scan lines—exceeds the film mode threshold 284 and deasserts (false) the film mode signal at 281. The deasserted film mode signal at 281 immediately switches the deinterlace apparatus to video mode allowing the switch 100 to select spatially interpolated, averaged, and temporal-spatially interpolated pixels at 71, 95, and 81, respectively, responsive to the motion and no motions signals at 251 and 263, respectively. The video mode will prevent the switch 100 from merging output pixels from the future or previous fields at 43 and 67, respectively.

[0047] FIG. 5 is a flowchart of an implementation of the spatial interpolator 70 shown in FIG. 1. Referring to FIGS. 1 and 5, the spatial interpolator 70 calculates a gradient for a plurality of directions from the current field sub image 50 provided at line 51. The plurality of directions might include the vertical (V, 90°), horizontal (H, 180°), diagonal (D, −45°), anti-diagonal (A, 45°), diagonal-1 (D1, −63°), anti-diagonal-1 (A1, 63°), diagonal-2 (D2, −27°), and anti-diagonal-2 (A2, 27°) directions. A person of reasonable skill in the art should understand directions other than those mentioned come within the scope of the invention.

[0048] FIG. 6A shows the eight directions we list above relative to the pixels in the current field, the pixels being represented by squares. FIG. 6B shows the eight directions relative to the pixel to be interpolated x, with the rows immediately above and below. The pixel to be interpolated x is at the crossing point of the eight directions. Referring to FIG. 6B, except for the horizontal H, diagonal-1 D1, antidiagonal-1 A1 directions, each line on any direction passes the center of one pixel on the row above and one pixel on the row below the pixel to be interpolated x. Lines D1 and A1 cross midway between two pixels on the row above and two pixels on the row below the pixel to be interpolated x. The horizontal line H does not pass any pixels in the current field as we expect.

[0049] Referring back to FIGS. 1 and 5, the gradient estimations for the eight directions are represented by G_v, G_h, G_d, G_a, G_d1, G_a1, G_d2 and G_a2. The spatial interpolator 70 calculates the eight gradients G_v, G_h, G_d, G_a, G_d1, G_a1, G_d2 and G_a2 at 702. We describe an embodiment of these calculations below. The spatial interpolator 70 determines the minimum gradient G_min at 704. The spatial interpolator 70 then compares each of the eight gradients G_v, G_h, G_d, G_a, G_d, G_a1, G_d2 and G_a2 against the minimum gradient G_min.

[0050] For example, at 706, the spatial interpolator 70 determines whether the minimum gradient G_min is equal to the vertical gradient G_v. If it is, it calculates the output pixel Pout at 710.

[0051] If it is not, the spatial interpolator 70 determines whether the minimum gradient G_min is equal to the horizontal gradient G_h at 716. If it is, it calculates the output pixel Pout at 720.

[0052] If it is not, the spatial interpolator 70 determines whether the minimum gradient G_min is equal to the diagonal-1 gradient G_d1 at 726. If it is, it calculates the output pixel Pout at 730.

[0053] If it is not, the spatial interpolator 70 determines whether the minimum gradient G_min is equal to the anti-diagonal-1 gradient G_a1 at 736. If it is, it calculates the output pixel Pout at 740.

[0054] If it is not, the spatial interpolator 70 determines whether the minimum gradient G_min is equal to the diagonal gradient G_d at 746. If it is, it calculates the output pixel Pout at 750.

[0055] If it is not, the spatial interpolator 70 determines whether the minimum gradient G_min is equal to the anti-diagonal gradient G_a at 756. If it is, it calculates the output pixel Pout at 760.

[0056] If it is not, the spatial interpolator 70 determines whether the minimum gradient G_min is equal to the diagonal-2 gradient G_d2 at 766. If it is, it calculates the output pixel Pout at 770.

[0057] If it is not, the spatial interpolator 70 determines the minimum gradient G_min is equal to the anti-diagonal-2 gradient G_a2 and calculates the output pixel Pout at 780.

[0058] FIGS. 7A–D are diagrams of a vertical line in the current field. The circle at the center of the diagram represents the pixel to be interpolated x (FIG. 3). The line shown in FIG. 7A might not be classified as a line even after interpolation because of its short length (3 pixels). The vertical lines shown in FIGS. 7B and 7C are likely the ends of vertical lines, but they might also represent lines with short lengths (3 pixels) in the current field only, or five or six pixels after interpolation. Put differently, the vertical lines shown in FIGS. 7B and 7C might represent very short lines. FIG. 7D presents a fourth case showing a high confidence vertical line, with a longer length of at least 7 pixels after interpolation.

[0059] FIGS. 8A–D are diagrams of a diagonal line in the current field. Similarly to the vertical lines shown in FIGS. 7A–D, the diagonal line of FIG. 8A might be too short to classify as a diagonal line. The diagonal lines shown in FIGS. 8B and 8C are likely the ends of diagonal lines. FIGS. 6A–B and FIGS. 7A–D show that the deinterlace apparatus needs about a two pixel line length in the current field to determine whether there is a line or edge passing through the pixel to be interpolated x (FIG. 3).

[0060] When a line or edge passes through the pixel to be interpolated x (FIG. 3), the immediately adjacent neighbor on either side of the line or edge will have almost identical properties to the line or edge. The gradient in the direction of the line or edge, therefore, will likely be the smallest. The pixels marked with an x in FIGS. 6A and 7A–D are the pixels adjacent to lines. To measure the gradient of the pixel in a particular direction, a weighted sum of the absolute differences of a surrounding neighborhood on a given direction is taken into account.

[0061] The gradient is the derivative of a function on a given direction. In the digital domain, the derivative is replaced by a difference. The gradient is thus measured by differentiation of variously spaced pixels. For example, referring to FIG. 3, assume the absolute difference of a single pair of pixels measures the gradient. And assume that |P24−P44|=|P23−P45|. Under these assumptions, we would not conclude that the gradient along the vertical direction is the same as the gradient along the diagonal direction. The gradient along the vertical direction is approximately P24–P44 divided by the distance between P24 and P44. Likewise, the gradient along the diagonal direction is approximately P23–P45 divided by the distance between P23 and P45. Since the distance between P23 and P45 is 1.41 times the distance between P24 and P44, the gradient along the vertical direction is 0.41 times greater than the gradient along the diagonal direction.

[0062] In one embodiment, the gradient is measured for the eight directions shown in FIG. 6. A person of skill in the art should recognize that other directions come within the scope of the present invention.

[0063] The gradient calculations for the eight directions are as follows.

Horizontal (H) direction

G_h=w_h[a₀(|P23−P24|+|P24−P25|+|P43−P44|+|P44−P45|)+a₁(|P22−P23|+|P25−P26|+|P42−P43|+|P45−P46|)]

Vertical (V) direction

G_v=w_v[b₀|P24−P44 |+b₁(|P23−P43|+|P25−P45|)+b₂(|PO4−P24|+|P44−P46|)+b₃(|PO3−P23|+|P43−P63|+|P05−P25|+|P45−P65|)]

Diagonal (D) direction

G_d=w_d[c₀|P23−P45|+c₁(|P22−P44|+|P24−P46|)+c₂(|P01−P23|+|P45−P67|+|P02−P24|+|P44−P66|)]

Anti-diagonal (A) direction

G_a=w_d[c₀|P25−P43|+c₁(|P24−P42|+|P26−P44|)+C₂(|PO7−P25|+|P43−P61|+|P06−P24|+|P44−P62|)]

Diagonal-1 (D1) direction

G_d1=w_d1[d₀(|P23−P44|+|P24−P45|)+d₁(|P22−P43|+|P25−P46|)+d₂(|PO3−P24|+|P44−P65|)+d₃(|PO2−P23|+|P45−P66|)]

Anti-diagonal-1 (A1) direction

G_a1=w_d1[d₀(|P25−P44|+|P24−P43|)+d₁(|P23−P42|+|P26−P45|)+d₂(|PO5−P24|+|P44−P63|)+d₃(|PO6−P25|+|P43−P62|)]

Diagonal-2 (D2) direction

Gd₂=w_d2[e₀|P22−P46|+e₁(|P21−P45|+|P23−P47|)+e₂(|P00−P24|+|P24−P48|+|P20−44|+|P44−P68|)]

Anti-diagonal-2 (A2) direction

G_a2=w_d2[e₀|P26−P42|+e₁(|P25−P41|+|P27−P43|)+e₂(|PO8−P24|+P24−P40|+P28−P44|+|P44−P60|)]

[0064] In the gradient calculations, w's, a₀, b₀, c₀, d₀, and e₀ are predetermined weights. The gradient for a plurality of directions, e.g., eight directions, will be calculated for the pixel to be interpolated x. The minimum gradient will give the interpolation direction.

[0065] We implement one embodiment of the spatial interpolator 70 as follows. We implement interpolation in the vertical, diagonal, and anti-diagonal directions using a four-tap low pass half-band filter. We implement interpolation in the diagonal-2 and anti-diagonal-2 directions using a two-tap low pass half-band filter. Horizontal interpolation is impossible since the horizontal line is entirely missing. Thus, we obtain the horizontal line by applying a low pass filter to the rows above and below and then averaging.

[0066] FIG. 5 lists the formulas we use to calculate the output pixels depending on the minimum gradient direction. For example, if the minimum gradient G_min is equal to the vertical gradient G_v at 706, we calculate the output pixel Pout with the formula given at 710. If the minimum gradient G_min is equal to the horizontal gradient G_h at 716, we calculate the output Pout with the formula given at 720. If the minimum gradient G_min is equal to the diagonal-1 gradient G_d1 at 726, we calculate the output Pout with the formula given at 730. If the minimum gradient G_min is equal to the anti-diagonal-1 gradient Gal at 736, we calculate the output Pout with the formula given at 740. If the minimum gradient G_min is equal to the diagonal gradient G_d at 746, we calculate the output Pout with the formula given at 750. If the minimum gradient G_min is equal to the diagonal gradient G_a at 756, we calculate the output Pout with the formula given at 760. If the minimum gradient G_min is equal to the diagonal-2 gradient G_d2 at 766, we calculate the output Pout with the formula given at 770. If the minimum gradient G_min is equal to the anti-diagonal-2 gradient G_a2, we calculate the output Pout with the formula given at 780.

[0067] FIG. 9 is a diagram of the interpolation method used in an embodiment of the present invention to obtain an output pixel along the diagonal D1 direction. In FIG. 9, the pixel to be interpolated x (FIG. 3) is shown as output pixel P34. Two intermediate pixels A and B are generated by linear interpolation along the D1 direction. Then, output pixel P34 is obtained by averaging the value of pixels A and B. Since line AB is perpendicular to direction D1, interpolation occurs on an orthogonal base.

[0068] From simple geometry, the ratio of the lengths {overscore (P23A)} and {overscore (AP44)} is 3:2. That yields the formulas given in 730 and 740. An embodiment of the present invention, therefore, provides a mean of interpolation along D1 direction without requiring super-sampling. The interpolation method of the present invention is also more accurate than methods requiring super-sampling.

[0069] FIG. 10 is a flowchart of an implementation of the spatial-temporal interpolator 80 shown in FIG. 1. Referring to FIGS. 1 and 5, the switch 100 selects output pixels generated by the spatial-temporal interpolator 80 when there exists uncertainty in the status of the pixel to be interpolated x (FIG. 3). That is, it is unclear whether the pixel to be interpolated x is part of a moving or still portion of the video image. Even if the pixel to be interpolated x belongs to a moving part of the video image, that motion is not strong. The current, previous, and/or future fields are therefore taken into account to interpolate the output pixel x.

[0070] In one embodiment of the spatial-temporal interpolator 80, we use four directions, e.g., vertical, horizontal, diagonal, and anti-diagonal to interpolate the pixel x. A person of reasonable skill in the art should understand directions other than those mentioned can be used instead of or in conjunction with those we mention above, and that all combinations thereof come within the scope of the invention.

[0071] Referring to FIGS. 1 and 10, the spatial-temporal interpolator 80 calculates a gradient for a plurality of directions from the current field sub image 50 provided at line 53 and the previous field sub image 62 provided at line 65 (802). In one embodiment, the spatial-temporal interpolator 80 calculates the gradients in the four directions, e.g., vertical (V), horizontal (H), diagonal (D), and anti-diagonal (A). It is worth noting that the gradient calculations are the same as those for the spatial interpolator 70. The gradient calculations, therefore, need only be performed once to be used by both the spatial interpolator 70 and the spatial-temporal interpolator 80.

[0072] The spatial-temporal interpolator 80 determines the minimum gradient G_min at 804.

[0073] The spatial-temporal interpolator 80 then compares each of the four gradients G_v, G_h, G_d, and G_a, against the minimum gradient G_min.

[0074] For example, at 806, the spatial-temporal interpolator 80 determines whether the minimum gradient G_min is equal to the vertical gradient G_v. If it is, it calculates the output pixel Pout at 810.

[0075] If it is not, the spatial-temporal interpolator 80 determines whether the minimum gradient G_min is equal to the horizontal gradient G_h at 816. If it is, it calculates the output pixel Pout at 820.

[0076] If it is not, the spatial-temporal interpolator 80 determines whether the minimum gradient G_min is equal to the diagonal gradient G_d at 826. If it is, it calculates the output pixel Pout at 830.

[0077] If it is not, the spatial-temporal interpolator 80 determines the minimum gradient G_min is equal to the anti-diagonal gradient G_a and calculates the output pixel Pout at 840.

[0078] We implement one embodiment of the spatial-temporal interpolator 80 as follows. We implement interpolation in the vertical, diagonal, and anti-diagonal directions using a seven-tap spatial-temporal filter. We obtain the horizontal line by averaging the still and moving image portions.

[0079] It should be readily apparent that the deinterlace method and apparatus might be implemented using hardware, software, firmware, or a combination thereof. The deinterlace method and apparatus might include a microprocessor or a microcontroller. And the deinterlace apparatus and method might be specially constructed for the required purposes or it might comprise a general-purpose computer selectively activated or reconfigured by a computer program stored in the computer.

[0080] We have described and illustrated the principles of our invention. Those of skill in the art should recognize that the invention might be modified in arrangement and detail without departing from these principles. We claim all modifications coming within the spirit and scope of the accompanying claims.