Method for motion estimation

[0001] This Non-provisional application claims priority under 35 U.S.C. § 119(a) on Patent Application No(s). 2004-102756 filed in Japan on Mar. 31, 2004, the entire contents of which are hereby incorporated by reference.

[0002] 1. Field of the Invention

[0003] The present invention relates to motion estimation and particularly to a spiderweb search method for motion estimation to identify the reference macro-block.

[0004] 2. Description of the Related Art

[0005] An MPEG video sequence is comprised of one or more groups of pictures, each group of which is composed of one or more pictures of type I-, P-, or B-. Intra-coded pictures, or “I-pictures,” are coded independently without reference to any other pictures. Predictive-coded pictures, or “P-pictures,” use information from preceding reference pictures, while bi-directionally predictive-coded pictures, or “B-pictures,” may use information from preceding or upcoming pictures, both, or neither.

[0006] Motion estimation is the process of estimating the displacement of a portion of an image between neighboring pictures. For example, a moving soccer ball will appear in different locations in adjacent pictures. Displacement is described as the motion vectors that give the best match between a specified region, e.g., the ball, in the current picture and the corresponding displaced region in a preceding or upcoming reference picture. The difference between the specified region in the current picture and the corresponding displaced region in the reference picture is referred to as “residue”.

[0007] In general, two known types of motion estimation methods used to estimate the motion vectors are pixel-recursive algorithms and block-matching algorithms. Pixel-recursive techniques predict the displacement of each pixel iteratively from corresponding pixels in neighboring frames. Block-matching algorithms, on the other hand, estimate the displacement between frames on a block-by-block basis and choose vectors that minimize the difference.

[0008] In conventional block-matching processes, the current image to be encoded is divided into equal-sized blocks of pixel information. In MPEG-1 and MPEG-2 video compression standards, for example, the pixels are grouped into “macro-blocks,” each consisting of a 16×16 sample array of luminance samples together with one 8×8 block of samples for each of the two chrominance components. The 16×16 array of luminance samples further comprises four 8×8 blocks that are typically used as input blocks to the compression models.

[0009] A block-matching algorithm that compares the current block to every candidate block within the search range is called a “full search”. In general, larger search areas generally produce a more accurate displacement vector, however, the computational complexity of a full search is proportional to the size of the search area and is too slow for some applications. A full search block-matching algorithm applied on a macro-block of size 16×16 pixels over a search range of ±N pixels with one pixel accuracy, for example, requires (2×N+1)₂ block comparisons. For N=16, 1089 16×16 block comparisons are required. Because each block comparison requires 16×16, or 256, calculations, this method is computationally intensive and operationally very slow. Techniques that simply reduce the size of the search area, however, run a greater risk of failing to find the optimal matching block.

[0010] As a result, there has been much emphasis on producing fast algorithms for finding the matching block within a wide search range. Most fast search techniques gain speed by computing the displacement only for a sparse sampling of the full search area. The 2-D logarithmic search, for example, reduces the number of computations by computing the MSE for sparsely-spaced candidates, and then successively searching the closer spaced candidates surrounding the best candidate found in the previous iteration. In a conjugate direction search, the algorithm searches in a horizontal direction until a minimum distortion is found. Then, proceeding from that point, the algorithm searches in a vertical direction until a minimum is found. Both of these methods are faster than a full search but frequently fail to locate the optimal matching block.

[0011] Other methods for overcoming the disadvantages of a full search have employed hierarchical search techniques. In a first stage, for example, a coarse search is performed over a reasonably large area. In successive stages of a conventional hierarchical search, the size of the search area is reduced. One example of a three-step hierarchical search is described in H. M. Jong et al., “Parallel Architectures for 3-Step Hierarchical Search Block-Matching Algorithm,” IEEE Trans. On Circuits and Systems for Video Technology, Vol. 4, August 1994, pp. 407–416. The hierarchical search described in Jong et al. is inadequate for some applications because the coarse search does not utilize all of the pixel information and thus may form an incorrect starting point for the finer search.

[0012] The object of the present invention is to provide a fast and effective method for motion estimation adopting a spider-web search method to identify a reference macro-block.

[0013] The present invention provides a method for motion estimation to determine a motion vector from a macro-block to be encoded to a reference macro-block. The method includes the steps: a) dividing a search area into sub-areas respectively having different densities, b) deriving distortions respectively between the macro-block to be encoded and a currently selected macro-block, and the macro-block to be encoded and neighbors of the currently selected macro-block, wherein the neighbors are determined according to the density of the sub-area in which the currently selected macro-block is located, c) if one of the distortions between the neighbors and the macro-block to be encoded is smallest, identifying the neighbor having the smallest distortion as the currently selected macro-block, d) repeating steps b) and c) until the currently selected macro-block has the smallest distortion among the neighbors thereof, e) if the currently selected macro-block having the smallest distortion is located a sub-area other than that having the largest density, re-defining the search area according to the density of the sub-area in which the currently selected macro-block is located, f) repeating steps a), b), c), d) and e) until the currently selected macro-block having the smallest distortion is located in the sub-area having the largest density, and g) identifying the currently selected macro-block having the smallest distortion and located in the sub-area having the largest density as the reference macro-block.

[0014] The present invention provides another method for motion estimation to determine a motion vector from a macro-block to be encoded to a reference macro-block. This method includes the steps: a) dividing a search area into sub-areas respectively having different densities, b) deriving distortions respectively between the macro-block to be encoded and a currently selected macro-block (X,Y), and the macro-block to be encoded and eight neighbors (X+D_m,Y−D_m), (X+D_m, Y), (X+D_m, Y+D_m), (X, Y−D_m), (X, Y+D_m), (X−D_m, Y−D_m), (X−D_m,Y), (X−D_m,Y+D_m) of the currently selected macro-block, wherein 1/D_m is the density of the sub-area in which the currently selected macro-block (X,Y) is located, c) if one of the distortions between the neighbors and the macro-block to be encoded is smallest, identifying the neighbor having the smallest distortion as the currently selected macro-block, d) repeating steps b) and c) until the currently selected macro-block has the smallest distortion among the neighbors thereof, e) if the currently selected macro-block having the smallest distortion is located in one of the sub-areas other than that having the largest density, re-defining the search area and dividing the re-defined search area into sub-areas respectively having densities 1/D₁, . . . , 1/D_p, wherein 1/D_m′ is the density of the sub-area in which the currently selected macro-block is located and D₁+ . . . +D_(p−1)≦D_m′≦D+ . . . +D_p, f) repeating steps b), c), d) and e) until the currently selected macro-block having the smallest distortion is located in the sub-area having the largest density, and g) identifying the currently selected macro-block having the smallest distortion and located in the sub-area having the largest density as the reference macro-block.

[0018] FIG. 1 is a flowchart of a method for motion estimation according to one embodiment of the invention. The object of the motion estimation is to determine a motion vector from a macro-block to be encoded to a reference macro-block. The encoded and reference macro-blocks may reside in two different frames. The steps shown in FIG. 1 are described with reference to examples shown in FIGS. 2A˜2I in the following.

[0019] In step S10, n discrete densities 1/D₁ ˜1/D_n are pre-determined, such as 4 densities of 1, ½, ⅓ and ¼.

[0020] In step S11, the search area is initially comprised of all macro-blocks in a frame and divided into sub-areas respectively having the densities 1/D₁˜1/D_n. The sub-areas are ring-like and concentric at an initially selected macro-block (x_i,y_i), and the density of a sub-area closer to the center macro-block is larger. More specifically, a sub-area having a density 1/D_m(1≦m≦n) comprises macro-blocks (X,Y) wherein X and Y conform to the following equations: $\hspace{1em}\{\begin{array}{c}\left[x_i+\left(D_1+\dots +D_{\left(m-1\right)}\right)+1\le X\le x_i+\left(D_1+\dots +D_m\right)\right]\mathrm{or}\left[x_i-\left(D_1+\dots +D_{\left(m-1\right)}\right)-1\ge X\ge x_i-\left(D_1+\dots +D_m\right)\right]\\ \left[y_i+\left(D_1+\dots +D_{\left(m-1\right)}\right)+1\le Y\le y_i+\left(D_1+\dots +D_m\right)\right]\mathrm{or}\left[y_i-\left(D_1+\dots +D_{\left(m-1\right)}\right)-1\ge Y\ge y_i-\left(D_1+\dots +D_m\right)\right]\end{array}$

[0021] As shown in FIG. 2A, the search area is a 21×21 array of macro-blocks and the macro-block (0,0) is initially selected. The search area is divided into four sub-areas respectively having densities 1, ½, ⅓ and ¼ such that, for example, the sub-area having the density ⅓ comprises the macro-blocks (X,Y) wherein 4≦X≦6 or −4≧X≧−6, and 4≦Y≦6 or −4≧Y≧−6. It should be noted that the initially selected macro-block is preferably located at a position in the referenced frame the same as that of the macro-block to be encoded in the encoded frame.

[0022] In step S12, distortions respectively between the macro-block to be encoded and a currently selected macro-block, and the macro-block to be encoded and neighbors of the currently selected macro-block are derived. The neighbors are determined according to the density of the sub-area in which the currently selected macro-block is located. More specifically, the neighbors of a currently selected macro-block (X,Y) in a sub-area having a density 1/D_m are located at (X+D_m,Y), (X−D_m,Y), (X+D_m,Y−D_m), (X,Y−D_m), (X−D_m,Y−D_m), (X+D_m,Y+D_m), (X, Y+D_m) and (X−D_m,Y+D_m). As shown in FIG. 2B, the distortions of the initially selected macro-block and its neighbors are derived. The initially selected macro-block is located in the sub-area having the density 1 so that its neighbors are located at (1,1), (1,0), (1,−1), (0,1), (0,−1), (−1,1), (−1,0) and (−1,−1). Each distortion is preferably a sum of absolute distance, a mean square error, or a mean absolute difference between two macro-blocks.

[0023] In step S13, it is determined whether the distortion between the currently selected macro-block and the macro-block to be encoded is smallest. If so, the procedure goes to step S15; otherwise, the procedure goes to step S14.

[0024] In step S14, the neighbor having the smallest distortion is identified as the currently selected macro-block. The procedure then returns to step S12. FIGS. 2C˜2F show the iterations of steps S12˜S14. As shown in FIG. 2C, the neighbor (−1,1) has the smallest distortion with the macro-block to be encoded and accordingly is identified as the currently selected macro-block. The newly selected macro-block (−1,1) is still located in the sub-area having the density 1 so that its neighbors are located at (−2,2), (−2,1), (−2,0), (−1,2), (−1,0), (0,2), (0,1) and (0,0). Then, as shown in FIG. 2D, the neighbor (−2,0) has the smallest distortion and accordingly is identified as the currently selected macro-block. The newly selected macro-block (−2,0) is now located in the sub-area having the density ½ so that its neighbors are located at (−4,2), (−4,0), (−4,−2), (−2,2), (−2,−2), (0,2), (0,0) and (0,−2). Next, as shown in FIG. 2E, the neighbor (−4,0) has the smallest distortion and accordingly is identified as the currently selected macro-block. The newly selected macro-block (−4,0) is located in the sub-area having the density ⅓ so that its neighbors are located at (−7,3), (−7,0), (−7,−3), (−4,3), (−4,−3), (−1,3), (−1,0) and (−1,−3). As shown in FIG. 2F, among the neighbors of the macro-block (−4,0), the macro-block (−7,−3) has the smallest distortion and accordingly is identified as the currently selected macro-block. The newly selected macro-block (−7,−3) is located in the sub-area having the density ¼ so that its neighbors are located at (−7,1), (−7,−7), (−3,1), (−3,−3) and (−3,−7). It should be noted that the macro-block (−7,−3) only has five neighbors since the others are out of the search area.

[0025] In step S15, it is further determined whether the currently selected macro-block having the smallest distortion is located in the sub-areas having the largest density. If so, the procedure goes to step S17; otherwise, the procedure goes to step S16.

[0026] In step S16, the search area is re-defined and re-divided according to the density of the sub-area in which the currently selected macro-block is located. More specifically, if the currently selected macro-block (x_c,y_c) is located in a sub-area having a density 1/D_m, the search area is re-defined as an area comprising macro-blocks (X,Y) wherein x_c−D_m≦X≦x_c+D_m and y_c−D_m≦Y≦y_c+D_m. The newly defined search area is divided into sub-areas having the densities 1/D₁˜1/D_p (1≦p≦n) in a manner the same as the step S11, wherein D₁+ . . . +D_(p−1)≦D_m≦D₁+ . . . +D_p. As shown in FIG. 2G, since the currently selected macro-block is located at (−7,−3) and in the sub-area having a density ¼, the search area is re-defined to comprise macro-blocks (X,Y) wherein −10≦X≦−3 and −7≦Y≦1. It is noted that the region (−11,Y) outside of the frame is excluded. Further, the newly defined search area is divided into three sub-areas respectively having densities 1, ½ and ⅓ since ≦14+≦21+2+3. After, the procedure returns to step S12. As shown in FIG. 2H, the current selected macro-block (−7,−3) is located in the new sub-area having the density 1 such that its neighbors are located at (−8,−4), (−8,−3), (−8,−2), (−7,−4), (−7,−2), (−6,−4), (−6,−3) and (−6,−2). Among the neighbors of the macro-block (−7,−3), the macro-block (−6,−4) has the smallest distortion with the macro-block to be encoded and is accordingly identified as the currently selected macro-block, as shown in FIG. 2I. Finally, the macro-block (−6,−4) has the smallest distortion and is located in the sub-area having the largest density 1, determined in the step S15. Consequently, the procedure goes to the step S17.

[0027] In step S17, the currently selected macro-block having the smallest distortion and located in the sub-area having the largest density is identified as the reference macro-block. As a result, the motion vector of the macro-block to be encoded is determined.

[0028] From simulation results, performance improvement is shown when adopting the spiderweb search method to every test video sequence. The spiderweb search method may be integrated into a diamond or 2D log search method to increase speed. The spiderweb search method reduces the number of checkpoints for reference macro-block and has minimal impact on PSNR degradation.

[0029] In the previously described method, it is noted that the distance between the neighbors are inversely proportional to the density of the region they are located in. The search area may not be rectangular. The number of the neighbors may be, for example, 4 and needs not to be 8.

[0030] In conclusion, the present invention provides a fast and effective method for motion estimation adopting a spiderweb search method to identify a reference macro-block. The spiderweb search is easily implemented or applied to other estimation algorithms to further reduce computational complexity without sacrificing compression quality.

[0031] The foregoing description of the preferred embodiments of this invention has been presented for purposes of illustration and description. Obvious modifications or variations are possible in light of the above teaching. The embodiments were chosen and described to provide the best illustration of the principles of this invention and its practical application to thereby enable those skilled in the art to utilize the invention in various embodiments and with various modifications as are suited to the particular use contemplated. All such modifications and variations are within the scope of the present invention as determined by the appended claims when interpreted in accordance with the breadth to which they are fairly, legally, and equitably entitled.