Pattern analysis-based motion vector compensation apparatus and method

• US 20050025342A1
• Filed: 07/12/2004
• Published: 02/03/2005
• Est. Priority Date: 07/31/2003
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: Lee, Sung-hee, Kwon, Ohjae
• Granted Patent: US 7,555,166 B2
• Time in Patent Office: Days
• US Class Current: 382/107
• CPC Class Codes: G06T 7/215 Motion-based segmentation
• CPC Class Codes: G06T 7/223 using block-matching
• CPC Class Codes: H04N 7/014 involving the use of motion vectors motion estimation and compensation in video coding H04N19/51

Abstract

Disclosed is an apparatus and method for compensating for estimated motion vectors by deciding whether a pattern image exists in target blocks for motion estimation. A motion estimation part compares a current block of a current frame/field to a search area established in a previous frame/field, calculates plural motion estimation error values, and estimates a temporary motion vector of the current block based on the plural motion estimation error values. A pattern decision part analyzes the calculated motion estimation error values, and decides that the current block is a pattern image. A motion compensation part produces a compensation motion vector of the current block by use of the plural motion estimation error values if the current block is decided as the pattern image. A motion vector selection part selects a final motion vector based on a result of the pattern decision part.

Claims

1. A pattern analysis-based motion vector compensation apparatus, comprising:

a motion estimation part for comparing a current block of a current frame or field with a certain search area established in a previous frame or field, calculating plural motion estimation error values, and estimating a temporary motion vector of the current block based on the plural motion estimation error values;

a pattern decision part for analyzing the calculated motion estimation error values, and deciding whether the current block is a pattern image;

a motion compensation part for producing a compensation motion vector of the current block by use of the plural motion estimation error values if the current block is decided as the pattern image; and

a motion vector selection part for selecting as a final motion vector either the temporary motion vector estimated by the motion estimation part or the compensation motion vector compensated for by the motion compensation part based on a result of the pattern decision part.

2. The pattern analysis-based motion vector compensation apparatus as claimed in claim 1, wherein the motion estimation part comprises:

a motion estimation error calculation unit for applying a block-matching approach to the current block and the search area of the current frame or field, and calculating the plural motion estimation error values; and

a motion vector estimation unit for estimating the temporary motion vector of the current block based on a point of a minimum motion estimation error value of the plural motion estimation error values produced.

3. The pattern analysis-based motion vector compensation apparatus as claimed in claim 1, wherein the pattern decision part comprises:

an integral projection unit for applying an integral projection to the plural motion estimation error values, and producing plural projection values;

a projection value verification unit for verifying the number of minimum local projection values between two points corresponding to predetermined first and second projection values but smaller than the first and second projection values of the plural projection values;

a period calculation unit for calculating the number of repeated periods including the at least one minimum local projection value sequentially appearing in the plural projection values, and dividing a length of the search area by the number of repeated periods to calculate the period; and

a pattern decision unit for analyzing the verified number of minimum local projection values and the calculated number of repeated periods, and deciding whether the current block is the pattern image.

4. The pattern analysis-based motion vector compensation apparatus as claimed in claim 3, wherein the integral projection unit comprises:

a vertical projector for applying the integral projection to the plural motion estimation error values in a vertical direction, and producing plural vertical projection values; and

a horizontal projector for applying the integral projection to the plural motion estimation error values in a horizontal direction, and producing plural horizontal projection values.

5. The pattern analysis-based motion vector compensation apparatus as claimed in claim 4, wherein the projection value verification unit comprises:

a vertical projection value verification portion for verifying the number of minimum vertical local projection values between two points corresponding to predetermined first and second vertical projection values but smaller than the first and second vertical projection values of the plural vertical projection values produced; and

a horizontal projection value verification portion for verifying the number of minimum local horizontal projection values between two points corresponding to predetermined first and second horizontal projection values but smaller than the first and second horizontal projection values of the plural horizontal projection values produced.

6. The pattern analysis-based motion vector compensation apparatus as claimed in claim 5, wherein the period calculation unit comprises:

a vertical period calculator for calculating the number of repeated vertical periods including at least one minimum local vertical projection values sequentially appearing in the plural vertical projection values; and

a horizontal period calculator for calculating the number of repeated horizontal periods including at least one or more minimum local horizontal projection values sequentially appearing in the plural horizontal projection values.

7. The pattern analysis-based motion vector compensation apparatus as claimed in claim 6, wherein the vertical and horizontal period calculators calculate the numbers of repeated vertical and horizontal periods based on the following equations:

8. The pattern analysis-based motion vector compensation apparatus as claimed in claim 6, wherein the vertical period calculator divides a horizontal length of the search area by the number of repeated vertical periods and calculates a vertical period of the pattern image, and the horizontal period calculator divides a vertical length of the search area by the number of repeated horizontal periods and calculates a horizontal period of the pattern image.

9. The pattern analysis-based motion vector compensation apparatus as claimed in claim 3, wherein the pattern decision part decides that the current block is a pattern image, if the number of the minimum local projection values is at least two and the number of repeated periods is larger than a predetermined first threshold value.

10. The pattern analysis-based motion vector compensation apparatus as claimed in claim 3, further comprising a pattern classification part for classifying the pattern image into a static pattern image having zero motion of the current block or into a dynamic pattern image occurring due to motion of the current block, depending on a point of a minimum local error value that is nearest at distance to the center of the search area if the current block is the pattern image.

11. The pattern analysis-based motion vector compensation apparatus as claimed in claim 10, wherein the pattern classification part compares a first minimum local error value located at a nearest distance to the center of the search area with the period calculated by the period calculation part, and classifies the pattern image, the first minimum local error value being one of at least one minimum local error values between two points corresponding to predetermined first and second error values but smaller than the first and second error values of the plural motion estimation error values.

12. The pattern analysis-based motion vector compensation apparatus as claimed in claim 11, wherein the pattern classification part decides that the pattern image is the static pattern image, if the pattern image satisfies a condition as below:

|u|=α

·

p

wherein, u denotes a point of the first minimum local error value that is located at the nearest distance to the center of the search area, a denotes a certain constant, and p denotes a period calculated by the period calculation part.

13. The pattern analysis-based motion vector compensation apparatus as claimed in claim 11, wherein the motion compensation part comprises:

a static motion compensation unit for estimating and producing the compensation motion vector based on a point corresponding to the first minimum local error value, if the pattern image is classified into the static pattern image;

an average vector calculation unit for calculating an average vector of motion vectors of blocks neighboring to the current block, if the pattern image is classified into the dynamic pattern image; and

a dynamic motion compensation unit for estimating and producing the compensation motion vector based on a point of a second minimum local error value of at least one or more minimum local error values that is located at a nearest distance to the calculated average vector.

14. The pattern analysis-based motion vector compensation apparatus as claimed in claim 1, wherein, if the pattern decision part decides that the current block is the pattern image, the motion vector selection part selects as the final motion vector the compensation motion vector estimated by the motion compensation part.

15. The pattern analysis-based motion vector compensation apparatus as claimed in claim 13, wherein, if the pattern classification part classifies the pattern image into the static pattern image, the motion vector selection part selects a compensation motion vector estimated by the static motion compensation unit, and, if the pattern classification part classifies the pattern image into the dynamic pattern image, the motion vector selection part selects a compensation motion vector estimated by the dynamic motion compensation unit.

16. The pattern analysis-based motion vector compensation apparatus as claimed in claim 1, wherein first and second motion estimation error values are calculated by use of Sum of Absolute Difference (SAD), Mean Absolute Difference(MAD), or Mean Square Error (MSE).

17. A pattern analysis-based motion vector compensation method, comprising:

a motion estimation step for comparing a current block of a current frame or field to a certain search area established in a previous frame or field, calculating plural motion estimation error values, and estimating a temporary motion vector of the current block based on the plural motion estimation error values;

a pattern decision step for analyzing the calculated motion estimation error values, and deciding whether the current block is a pattern image;

a motion compensation step for producing a compensation motion vector of the current block by use of the plural motion estimation error values if the current block is decided as the pattern image; and

a motion vector selection step for selecting as a final motion vector either the temporary motion vector estimated by the motion estimation step or the compensation motion vector compensated for by the motion compensation step based on a result of the pattern decision step.

18. The pattern analysis-based motion vector compensation method as claimed in claim 17, wherein the motion estimation step comprises:

a motion estimation error calculation step for applying a block-matching approach to the current block and the search area of the current frame or field, and calculating the plural motion estimation error values; and

a motion vector estimation step for estimating the temporary motion vector of the current block based on a point of a minimum motion estimation error value of the plural motion estimation error values produced.

19. The pattern analysis-based motion vector compensation method as claimed in claim 17, wherein the pattern decision step comprises:

an integral projection step for applying an integral projection to the plural motion estimation error values, and producing plural projection values;

a projection value verification step for verifying the number of minimum local projection values between two points corresponding to predetermined first and second projection values but smaller than the first and second projection values of the plural projection values;

a period calculation step for calculating the number of repeated periods including the at least one minimum local projection value sequentially appearing in the plural projection values, and dividing a length of the search area by the number of repeated periods to calculate the period; and

a pattern decision step for analyzing the verified number of minimum local projection values and the calculated number of repeated periods, and deciding whether the current block is the pattern image.

20. The pattern analysis-based motion vector compensation method as claimed in claim 19, wherein the integral projection step comprises:

a vertical projection step for applying the integral projection to the plural motion estimation error values in a vertical direction, and producing plural vertical projection values; and

a horizontal projection step for applying the integral projection to the plural motion estimation error values in a horizontal direction, and producing plural horizontal projection values.

21. The pattern analysis-based motion vector compensation method as claimed in claim 20, wherein the projection value verification step comprises:

a vertical projection value verification step for verifying the number of minimum vertical local projection values between two points corresponding to predetermined first and second vertical projection values but smaller than the first and second vertical projection values of the plural vertical projection values produced; and

a horizontal projection value verification step for verifying the number of minimum local horizontal projection values between the two points corresponding to predetermined first and second horizontal projection values but smaller than the first and second horizontal projection values of the plural horizontal projection values produced.

22. The pattern analysis-based motion vector compensation method as claimed in claim 21, wherein the period calculation method comprises:

a vertical period calculation step for calculating the number of repeated vertical periods including at least one minimum local vertical projection values sequentially appearing in the plural vertical projection values; and

a horizontal period calculation step for calculating the number of repeated horizontal periods including at least one minimum local horizontal projection values sequentially appearing in the plural horizontal projection values.

23. The pattern analysis-based motion vector compensation method as claimed in claim 22, wherein the vertical and horizontal period calculation steps calculate the numbers of repeated vertical and horizontal periods based on the following equations:

24. The pattern analysis-based motion vector compensation method as claimed in claim 22, wherein the vertical period calculation step divides a horizontal length of the search area by the number of repeated vertical periods and calculates a vertical period of the pattern image, and the horizontal period calculation step divides a vertical length of the search area by the number of repeated horizontal periods and calculates a horizontal period of the pattern image.

25. The pattern analysis-based motion vector compensation method as claimed in claim 19, wherein the pattern decision step decides that the current block is a pattern image, if the number of the minimum local projection values is at least two and the number of repeated periods is larger than a predetermined first threshold value.

26. The pattern analysis-based motion vector compensation method as claimed in claim 19, further comprising a pattern classification step for classifying the pattern image into a static pattern image having zero motion of the current block or into a dynamic pattern image occurring due to motion of the current block, depending on a point of a minimum local error value that is located at a nearest distance to the center of the search area if the current block is the pattern image.

27. The pattern analysis-based motion vector compensation method as claimed in claim 26, wherein the pattern classification step compares a first minimum local error value located at the nearest distance to the center of the search area with the period calculated by the period calculation step, and classifies the pattern image, the first minimum local error value being one of at least one minimum local error values between two points corresponding to predetermined first and second error values but smaller than the first and second error values of the plural motion estimation error values.

28. The pattern analysis-based motion vector compensation method as claimed in claim 27, wherein the pattern classification step decides that the pattern image is the static pattern image, if the pattern image satisfies the following condition:

|u|=α

·

p

wherein, u denotes a point of the first minimum local error value that is nearest at distance, a denotes a certain constant, and p denotes a period calculated by the period calculation step.

29. The pattern analysis-based motion vector compensation method as claimed in claim 27, wherein the motion compensation step comprises:

a static motion compensation step for estimating and producing the compensation motion vector based on a point corresponding to the first minimum local error value, if the pattern image is classified into the static pattern image;

an average vector calculation step for calculating an average vector of motion vectors of blocks neighboring the current block, if the pattern image is classified into the dynamic pattern image; and

a dynamic motion compensation step for estimating and producing the compensation motion vector based on a point of a second minimum local error value of at least one or more minimum local error values that is located at a nearest distance to the calculated average vector.

30. The pattern analysis-based motion vector compensation method as claimed in claim 17, wherein, if the pattern decision step decides that the current block is the pattern image, the motion vector selection step selects as the final motion vector the compensation motion vector estimated by the motion compensation step.

31. The pattern analysis-based motion vector compensation method as claimed in claim 29, wherein, if the pattern classification step classifies the pattern image into the static pattern image, the motion vector selection step selects a compensation motion vector estimated by the static motion compensation step, and, if the pattern classification step classifies the pattern image into the dynamic pattern image, the motion vector selection step selects a compensation motion vector estimated by the dynamic motion compensation step.

32. The pattern analysis-based motion vector compensation method as claimed in claim 17, wherein first and second motion estimation error values are calculated by use of Sum of Absolute Difference (SAD), Mean Absolute Difference (MAD), or Mean Square Error (MSE).

33. A pattern analysis-based motion vector compensation apparatus, comprising:

a motion estimation part for comparing a current block of a current frame or field to a search area established in a previous frame or field, calculating plural motion estimation error values, and estimating a temporary motion vector of the current block based on the plural motion estimation error values;

a pattern decision part for analyzing the calculated motion estimation error values and deciding that the current block is a pattern image, and applying an integral projection to the motion estimation error values and verifying information of points of plural integral vectors produced;

a motion compensation part for producing a compensation motion vector of the current block by use of the information of the points of the plural integral vectors if the current block is decided as the pattern image; and

a motion vector selection part for selecting as a final motion vector either the temporary motion vector estimated by the motion estimation part or the compensation motion vector compensated for by the motion compensation part based on a result of the pattern decision part.

34. The pattern analysis-based motion vector compensation apparatus as claimed in claim 33, wherein the pattern decision part comprises:

an integral projection unit for applying an integral projection to the plural motion estimation error values, and producing plural projection values;

a projection value verification unit for verifying the number of at least one or more minimum local projection values having motion estimation error values smaller than motion estimation error values corresponding to predetermined first and second projection values of the plural projection values;

a period calculation unit for calculating the number of repeated periods including at least one minimum local projection value sequentially appearing in the plural projection values, and dividing a length of the search area by the number of repeated periods to calculate a period; and

a pattern decision unit for analyzing the verified number of minimum local projection values and the calculated number of repeated periods, and deciding whether the current block is the pattern image.

35. The pattern analysis-based motion vector compensation apparatus as claimed in claim 34, further comprising a pattern classification part for classifying the pattern image into a static pattern image having zero motion of the current block or into a dynamic pattern image occurring due to motion of the current block, based on a point of a minimum local error value that is located at a nearest distance to the center of the search area if the current block is the pattern image.

36. The pattern analysis-based motion vector compensation apparatus as claimed in claim 33, wherein, if the pattern decision part decides that the current block is the pattern image, the motion vector selection part selects as the final motion vector the compensation motion vector estimated by the motion compensation part.