GB2323993A

GB2323993A - Method for motion estimation in a moving picture

Info

Publication number: GB2323993A
Application number: GB9727156A
Authority: GB
Inventors: Kye-Ho Park; Hak-Soo Kim
Original assignee: Samsung Electronics Co Ltd
Current assignee: Samsung Electronics Co Ltd
Priority date: 1997-03-31
Filing date: 1997-12-24
Publication date: 1998-10-07
Anticipated expiration: 2017-12-24
Also published as: KR19980075571A; GB9727156D0; DE19756224A1; DE19756224C2; CN1174635C; JPH10341441A; CN1195255A; MY115706A; KR100265721B1; JP3268258B2; GB2323993B

Abstract

A method for estimating the motion within a moving picture using a two-dimensional triangle-patch wireframe model, comprising the steps of dividing (200) the current frame of an input image signal into N 1 N 2 macroblock units, dividing (202) each of the macroblocks into two to create triangle patches, calculating (204) pixel differences between a reference block and search blocks of the previous frame in a predetermined range on the basis of the coordinate values of the respective apexes of the triangle patches, and calculating (206) a motion vector of the reference block of the current frame from the coordinate value of the matching block of the previous frame having a minimum pixel difference value among the calculated pixel differences. Thus, it is possible to obtain a picture of good quality compared with a picture reconstructed by the conventional BMA method for estimating the motion.

Description

2323993 METHOD FOR MOTION ESTIMATION IN A MOVING PICTURE The present

invention relates to a method for estimating the motion of a moving picture, and more particularly, to a method for estimating the motion of a moving picture using a two-dimensional triangle-patch wireframe model.

In general, in a moving picture compression system, it is possible to reduce the number of bits consumed during an encoding procedure by removing redundancy in a moving picture. Therefore, in almost all moving picture compression systems such as MPEG H. 261, a method for estimating the motion is used in order to remove is redundancy in a moving picture. Motion estimation of the moving picture is performed by the following steps. The current frame is divided into many small reference blocks. Each reference block is compared with various search blocks of a previous frame. A difference degree, to which the search blocks of the previous frame are different from the reference blocks is calculated. The search block whose difference degree is smallest (hereinafter referred to as a matching block) is found for each reference block. A motion vector is estimated by the difference of coordinates between each reference block and the respective matching block. Meanwhile, the current frame is encoded by only the difference between pixels of the matching blocks and of the respective reference blocks having the above motion vector. The frame encoded as above is restored to the original image by combining each encoded pixel difference value and the respective pixel value of the matching block.

Figure 1 explains a conventional method for estimating a motion. The standard method for estimating 2 motion used in a conventional encoder is a block matching algorithm (hereinafter, BMA). Reference numerals 10 and 12 respectively denote the reference block of the current frame and the search block of the previous frame.

Referring to Figure 1, a motion estimation method using the BMA determines from which block of the previous frame K the reference block of the current frame K+1 came. In this method, a measuring principle such as a mean square error (MSE) represented by the following formula 1 or a minimum mean absolute difference (MAD) represented by the following formula 2 is used. The calculation based on such a measuring principle is performed on one block unit, in which the pixel differences between the reference block of the current frame and the search block of the previous frame are calculated. At this time, the search block giving the minimum value of the calculated difference values becomes the matching block and the motion is estimated by calculating a motion vector of the matching block. Namely, the motion vector is obtained by calculating the difference between the coordinates of the reference block and of the matching block.

[FORMULA 11 MSE (dl, d2) 1 E [ fkl (nl, n2) - f,,, (nl + dj, n2 + d2)l N1 N2(n,, n.) EB [FORMULA 21 mAD A, d2) = 1 E 1 fk., (nl, n2) - f. (n, + dj, n2 + d2H 1V1 1V2 (n,, n2) EB wherein, B=NIxN2.

Cl However, the conventional BMA method for estimating motion, it is hard to ref lect a rotating movement or a size-varied motion since it is created under the assumption that the motion is planar during estimation of the motion of a continuous moving picture. Thus, it is hard to estimate a subtle motion. Also, the picture quality is deteriorated when the encoded image is restored since a discontinuity of the image is generated between the blocks.

It is an aim of at least the preferred embodiments of the present invention to provide an improved method for estimating motion within a moving picture.

According to one aspect of the present invention, there is provided a method for estimating motion in a moving picture, said method for use in a moving picture compression system, said method comprising the steps of:

dividing a current frame of an input image signal into NIxN2 macroblock units; dividing each of said macroblocks into two to create triangle patches; calculating pixel differences between a reference block and search blocks of a previous frame in a predetermined range on the basis of the coordinate values of respective apexes of said triangle patches; and calculating a motion vector of the reference block of the current frame from the coordinate value of a matching 4 - block of the previous frame having a minimum pixel difference value among said calculated pixel differences.

Advantageously, the coinciding apexes of the triangles are calculated only once and the apexes on the border of the reference block are not calculated, in the pixel difference calculating step.

The pixel difference is preferably calculated as the absolute value of the difference between the pixel value of the current frame and the pixel value of the previous frame in the pixel difference calculating step.

The step of estimating the motion of a midway-pixel unit preferably comprises estimating a motion in more detail, on the basis of the calculated motion vector.

The midway-pixel value is preferably generated by the combination of peripheral full-pixel values.

For a better understanding of the invention, and to show how embodiments of the same may be carried into effect, reference will now be made, by way of example, to the accompanying diagrammatic drawings, in which:

Figure 1 explains a conventional method for estimating motion; Figure 2 explains a method for estimating the motion 30 of a moving picture using a two-dimensional triangle-patch wireframe model; Figure 3 explains a step of initializing the twodimensional triangle- patch wireframe model; - 5 Figure 4 explains a method for determining the position of a midway-pixel; and Figures SA through SC show examples of estimating the motion of a moving picture.

Referring firstly to Figure 2, a preferred method for estimating the motion of a moving picture is shown, using a two-dimensional trianglepatch wireframe model. The method includes steps for initializing the two-dimensional triangle-patch wireframe model and steps for estimating a motion vector from the initialized triangle-patch wireframe model.

First, the steps of initializing the two-dimensional triangle-patch wireframe model are as follows.

The current frame of an input image signal is divided into Nix N2 macroblock units (step 200). A two-dimensional triangle-patch is generated by dividing each macroblock into two (step 202). Namely, the triangle -patches are generated by drawing a diagonal line from the top left corner to the bottom right corner or from the top right corner to the bottom left corner of the macroblock as shown in Figure 3.

The steps for estimating the motion vector from the initialized trianglepatch wireframe model are as follows.

Pixel differences between a reference block of a predetermined size (for example, 16x 16 pixels), selected on the basis of the apexes of the generated triangle, and the search blocks of the previous frame, are respectively calculated (step 204). Namely, after setting the reference block of a predetermined size on the basis of the apexes of the triangle-patch, the pixel differences between the respective search blocks and the reference block are calculated within a predetermined search range ( 15 pixels). For example, the pixel difference values between all the search blocks having a predetermined search range such as +15 pixels and the reference block are calculated. At this time, coinciding triangle apexes are calculated only once and the apexes positioned on the border of the image are not calculated.

The coordinate value of the search block of the previous frame having a minimum pixel difference value from among all the pixel differences between the reference block and the search blocks, is generated as the motion 15 vector of the reference block of the current frame (step 206). Here, the motion vector (Mv,,MVY) is calculated by a sum of difference (SAD) represented by the following formula 3.

[FORMULA 31 N, N SAD,, W, v) = E 1 rb (i, j) - sw (i + u, j + v) i=i, j=I wherein, N is equal to 16, rb(i,j) is the (i,j)th pixel value in the reference block of the current frame, and sw(i,j) is the (i,j)th pixel value in the search block of the previous frame.

Meanwhile, SAD(0,0) value for a zero vector is subtracted from the predetermined value (for example, 100), and the result is compared with the SAD values of other vectors, that is to say, the SAD value for the zero vector has little difference from the SAD values for the motion vectors of another search block, the zero vector is - 7 given priority in determining as a moving vector. Thus, the number of bits consumed on encoding the motion vector are reduced. Then, the minimal SAD value among SAD values for each (x,y) is searched, the (x,y) having the minimal SAD value is selected as a moving vector (MVxf MVY) [FORMULA 41 SAD (0, 0) =SAD (0, 0) - 10 0 Furthermore, an additional step of minutely estimating the motion is performed using the motion vector estimated through the above-mentioned step. The motion vector in a midway-pixel unit is searched for in a predetermined search range (for example, +6 pixels) as shown in Figure 4 on the basis of the motion vector extracted through the above-mentioned step. Namely, the coordinate values (+) of the pixels marked with A, B, C and D shown in Figure 4 denote the coordinate values of the pixels extracted through the above mentioned step. The coordinate values (0) of the pixels marked with a, b, c and d denote the coordinate values of the midway-pixels, which are generated by the combination of the coordinate values marked with capital letters. Here, the coordinate values of the midway-pixels are generated by the following formula 5.

[FORMULA 51 a=A, b=(A+B)/2 c=(A+C)/2, d= (A+B+C+D) /4 At this time, the motion vectors (Mvx,MVy) of the pixels with respect to the respective coordinate values are renewed. A final motion vector is estimated from the renewed pixels using the formulas 3 and 4.

8 Figures SA through 5C show examples of applying the present invention to estimate the motion of a moving picture. Figure 5A shows the image of the previous frame. Figure 5B shows the image of the current frame. Figure 5C shows the image transformed after motion estimation using a twodimensional triangle-patch wireframe model. It is noted from Figure 5C that a discontinuity is not generated on the borders of the respective triangles when the image is reconstructed in a unit of pixels in the respective triangles during motion compensation since the respective triangles in the two-dimensional triangle-patch wireframe model are connected to each other after the estimation of the motion.

is When the image encoded by the method for estimating the motion of the moving picture using the two-dimensional triangle-patch wireframe model is reconstructed, it is possible to obtain a picture of good quality compared with an image reconstructed by the conventional BMA method for estimating the motion.

The reader's attention is directed to all papers and documents which are filed concurrently with or previous to this specification in connection with this application and which are open to public inspection with this specification, and the contents of all such papers and documents are incorporated herein by reference.

All of the features disclosed in this specification (including any accompanying claims, abstract and drawings), and/or all of the steps of any method or process so disclosed, may be combined in any combination, except combinations where at least some of such features and/or steps are mutually exclusive.

- 9 Each feature disclosed in this specification (including any accompanying claims, abstract and drawings), may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise. Thus, unless expressly stated otherwise, each feature disclosed is one example only of a generic series of equivalent or similar features.

The invention is not restricted to the details of the foregoing embodiment (s). The invention extends to any novel one, or any novel combination, of the features disclosed in this specification (including any accompanying claims, abstract and drawings), or to any novel one, or any novel combination, of the steps of any method or process so disclosed.

- 10

Claims

1. A method for estimating motion in a moving picture, said method for use in a moving picture compression system, said method comprising the steps of:

dividing a current frame of an input image signal into NIxN2 macroblock units; dividing each of said macroblocks into two to create triangle patches; calculating pixel differences between a reference block and search blocks of a previous frame in a predetermined range on the basis of the coordinate values of respective apexes of said triangle patches; and calculating a motion vector of the reference block of the current frame from the coordinate value of a matching block of the previous frame having a minimum pixel difference value among said calculated pixel differences.

2. The method of claim 1, wherein the coinciding apexes of said triangles are calculated only once and the apexes 2S on the border of said reference block are not calculated, in said pixel difference calculating step.

The method of claim 1 or 2, wherein the pixel difference is calculated as the absolute value of the difference between the pixel value of the current frame and the pixel value of the previous frame in said pixel difference calculating step.

4. The method of claim 1, 2 or 3, further comprising the step of estimating the motion of a midway-pixel unit for 11 estimating a motion in more detail, on the basis of said calculated motion vector.

5. The method of claim 4, wherein said midway-pixel value is generated by the combination of peripheral full-pixel values.

6. A method substantially as hereinbefore described with reference to figures 2 to 5 of the accompanying drawings.