# US 2003/0067987 A1 · Method for motion estimation in video coding

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2003/0067987 A1
- Document ID: US20030067987A1
- Status: Published application
- Kind: A1
- Application number: 10/241,593
- Filed: 2002-09-12
- Published: 2003-04-10
- Earliest claimed: 2001-10-09
- Assignees: Vivotek Inc.
- Inventors: Shyh-Yih Ma
- Web page: https://patentgrep.com/patent/US20030067987A1

## CPC

- H04N 5/145: {Movement estimation (for video coding H04N19/51)}
- G06T 7/238: using non-full search, e.g. three-step search
- H04N 19/51: Motion estimation or motion compensation
- H04N 19/557: Motion estimation characterised by stopping computation or iteration based on certain criteria, e.g. error magnitude being too large or early exit
- H04N 19/56: Motion estimation with initialisation of the vector search, e.g. estimating a good candidate to initiate a search
- H04N 19/583: Motion compensation with overlapping blocks
- H04N: PICTORIAL COMMUNICATION, e.g. TELEVISION
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

A method for motion estimation in video coding. The row containing a motion vector predictor is determined as a P-th row and the rows adjacent to the P-th row are the (P+1)-th row and the (P−1)-th row. The method includes the following steps: (a) calculating of the distortions corresponding to all points in the P-th, (P+1)-th, and (P−1)-th rows respectively; (b) comparing all distortions, (c) determining whether or not the point with a minimum distortion is in the P-th row, and if yes, the method proceeds to step (g); (d) determining whether or not the point with a minimum distortion is in the (P+1)-th row, and if yes, the method proceeds to step (f); (e) decreasing P by 1, calculating the corresponding distortions of all points in the (P−1)-th row respectively, and repeating step (b); (f) increasing P by 1, calculating the corresponding distortions of all points in the (P+1)-th row respectively, and repeating step (b); and (g) determining a corresponding motion vector from the point with a minimum distortion.

## Claims

### Claim 1

A method for motion estimation in video encoding for determining a motion vector corresponding to a first block in a current frame and a second block in a reference frame, wherein the second block in the reference frame is the best-matching block related to the first block in the current frame, a search region is predetermined and has a plurality of successive rows, including a (P-T)-th row, a P-th row, and a (P+1)-th row, each row of the search region has a plurality of points, and a motion vector predictor is located on the P-th row, the method comprising the steps of:
(a) calculating the respective distortions corresponding to all points in the P-th, (P+1)-th, and (P−1)-th rows respectively;
(b) comparing all distortions;
(c) determining whether or not the point with a minimum distortion is in the P-th row, the uppermost or the lowest row of the search region, and if yes, then the method proceeds to step (g);
(d) determining whether or not the point with a minimum distortion is in the (P+1)-th row, and if yes, then the method proceeds to step (f);
(e) decreasing P by 1, calculating the corresponding distortions of all points in the (P−1)-th row respectively, and repeating step (b);
(f) increasing P by 1, calculating the corresponding distortions of all points in the (P+1)-th row respectively, and repeating step (b);
(g) determining a corresponding motion vector from the point with a minimum distortion.

### Claim 2 (depends on 1)

A method according to claim 1, wherein the motion vector predictor is determined from the median of multiple motion vectors of multiple blocks adjacent to the first block.

### Claim 3 (depends on 1)

A method according to claim 1, wherein the distortion is the sum of absolute difference (SAD).

### Claim 4

A method for motion estimation in video encoding, for determining a motion vector corresponding to a first block in a current frame and a second block in a reference frame, wherein the second block in the reference frame is the best-matching block related to the first block in the current frame, a search region is predetermined and has a plurality of successive rows, including a (P−1)-th row, a P-th row, and a (P+1)-th row, each row of the search region has a plurality of points, and a motion vector predictor is located on the P-th row, the method comprising the steps of:
(a) calculating the distortions corresponding to all points in the P-th, (P+1)-th, and (P−1)-th rows respectively, wherein the distortion values corresponding to all points in the (P−1)-th row do not need to be calculated if the (P−1)-th row is not within the search region, and the distortion values corresponding to all points in the (P+1)-th row do not need to be calculated if the (P+1)-th is not within the search region;
(b) comparing all distortions;
(c) determining whether or not the point with a minimum distortion is in the P-th row, and if yes, then the method proceeds to step (g);
(d) determining whether or not the point with a minimum distortion is in the (P+1)-th row, and if yes, then the method proceeds to step (f);
(e) decreasing P by 1, calculating the corresponding distortions of all points in the (P−1)-th row respectively, and repeating step (b);
(f) increasing P by 1, calculating the corresponding distortions of all points in the (P+1)-th row respectively, and repeating step (b); and
(g) determining a corresponding motion vector from the point with a minimum distortion.

### Claim 5 (depends on 4)

A method according to claim 4, wherein the motion vector predictor is determined from the median of multiple motion vectors of multiple blocks adjacent to the first block.

### Claim 6 (depends on 4)

A method according to claim 4, wherein the distortion is the sum of absolute difference (SAD).
