Method and apparatus for fast motion estimation
Embodiments of the invention generally provide a method and apparatus for fast motion estimation. One embodiment of a method for motion estimation includes classifying a macroblock of a source frame in accordance with an initial matching cost, an intermediate matching cost, and a final matching cost, where the intermediate matching cost is estimated based on a predicted motion vector for the source frame and the final matching cost is estimated based on a final motion vector for a correlated macroblock in a prior frame, performing motion estimation for the macroblock based upon the classifying, wherein an early termination is applied in accordance with a result of the classifying, and outputting a motion vector for the macroblock based on a result of the motion estimation.
1. A method for motion estimation, comprising:
classifying a macroblock of a source frame into classes in accordance with an initial matching cost, an intermediate matching cost, and a final matching cost, where the intermediate matching cost is estimated based on a predicted motion vector for the source frame and the final matching cost is estimated based on a final motion vector for a correlated macroblock in a prior frame;
performing motion estimation for the macroblock based upon the classifying, wherein an early termination is applied in accordance with a result of the classifying, wherein the classifying classifies the macroblock of the source frame into one of three potential classes; and
outputting a motion vector for the macroblock based on a result of the motion estimation.
2. The method of claim 1 , wherein the three potential classes comprise:
a first class defined by macroblocks whose initial matching cost is less than a first predefined threshold;
a second class defined by macroblocks whose initial matching cost is greater than or equal to the first predefined threshold, and for whom an absolute value of a difference between the intermediate matching cost and the final matching cost is greater than a second predefined threshold; and
a third class defined by macroblocks whose initial matching cost is greater than or equal to the first predefined threshold, and for whom an absolute value of a difference between the intermediate matching cost and the final matching cost is less than or equal to a second predefined threshold.
3. The method of claim 2 , wherein the performing applies the early termination if the macroblock of the source frame is classified in the first class or the third class.
4. The method of claim 3 , wherein the early termination comprises ceasing the motion estimation after a small local search.
5. The method of claim 2 , wherein the first predefined threshold is based on a sum of absolute differences.
6. The method of claim 2 , further comprising:
computing the absolute value of the difference between the intermediate matching cost and the final matching cost;
classifying the macroblock of the source frame in the second class if the absolute value of the difference between the intermediate matching cost and the final matching cost is greater than the second predefined threshold; and
classifying the macroblock of the source frame in the third class if the absolute value of the difference between the intermediate matching cost and the final matching cost is less than or equal to the second predefined threshold.
7. The method of claim 6 , wherein the second predefined threshold is based on a motion vector amplitude.
8. The method of claim 7 , wherein the macroblock of the source frame is classified in accordance with:
Class
=
{
2
if
P
M
V
-
F
M
V
prior
>
t
3
if
P
M
V
-
F
M
V
prior
≤
t
where PMV is a predicted motion vector for the macroblock of the source frame and FMV prior is a motion vector of a macroblock in a frame immediately preceding the source frame that is correlated to the macroblock of the source frame, and t is the second predefined threshold.
9. The method of claim 1 , wherein the predicted motion vector is predicted by:
identifying two or more neighboring macroblocks that neighbor the macroblock of the source frame in the source frame;
computing a median of motion vectors for the two or more neighboring macroblocks; and
setting the median as the predicted motion vector for the macroblock of the source frame.
10. The method of claim 1 , wherein the initial matching cost is a metric based on a sum of absolute differences.
11. A non-transitory computer readable medium containing an executable program for motion estimation, where the program performs:
classifying a macroblock of a source frame into classes in accordance with an initial matching cost, an intermediate matching cost, and a final matching cost, where the intermediate matching cost is estimated based on a predicted motion vector for the source frame and the final matching cost is estimated based on a final motion vector for a correlated macroblock in a prior frame;
performing motion estimation for the macroblock based upon the classifying, wherein an early termination is applied in accordance with a result of the classifying, wherein the classifying classifies the macroblock of the source frame into one of three potential classes; and
outputting a motion vector for the macroblock based on a result of the motion estimation.
12. The non-transitory computer readable medium of claim 11 , wherein the classifying comprises:
classifying the macroblock of the source frame in a first class if the initial matching cost is less than a first predefined threshold;
classifying the macroblock of the source frame in a second class if the initial matching cost is greater than or equal to the first predefined threshold, and if an absolute value of a difference between an intermediate matching cost and a final matching cost is greater than a second predefined threshold; and
classifying the macroblock of the source frame in a third class if the initial matching cost is greater than or equal to the first predefined threshold, and if an absolute value of a difference between an intermediate matching cost and a final matching cost is less than or equal to a second predefined threshold.
13. The non-transitory computer readable medium of claim 12 , wherein the performing applies the early termination if the macroblock of the source frame is classified in the first class or the third class.
14. The non-transitory computer readable medium of claim 13 , wherein the early termination comprises ceasing the motion estimation after a small local search.
15. The non-transitory computer readable medium of claim 12 , wherein the first predefined threshold is based on a sum of absolute differences.
16. The non-transitory computer readable medium of claim 12 , further comprising:
computing the absolute value of the difference between the intermediate matching cost and the final matching cost;
classifying the macroblock of the source frame in the second class if the absolute value of the difference between the intermediate matching cost and the final matching cost is greater than the second predefined threshold; and
classifying the macroblock of the source frame in the third class if the absolute value of the difference between the intermediate matching cost and the final matching cost is less than or equal to the second predefined threshold.
17. The non-transitory computer readable medium of claim 16 , wherein the second predefined threshold is based on a motion vector amplitude.
18. The non-transitory computer readable medium of claim 17 , wherein the macroblock of the source frame is classified in accordance with:
Class
=
{
2
if
P
M
V
-
F
M
V
prior
>
t
3
if
P
M
V
-
F
M
V
prior
≤
t
where PMV is a predicted motion vector for the macroblock of the source frame and FMV prior is a motion vector of a macroblock in a frame immediately preceding the source frame that is correlated to the macroblock of the source frame, and t is the second predefined threshold.
19. A video encoder, comprising:
an input for receiving a source frame; and
a motion estimator for classifying a macroblock of the source frame into classes in accordance with an initial matching cost, an intermediate matching cost, and a final matching cost, where the intermediate matching cost is estimated based on a predicted motion vector for the source frame and the final matching cost is estimated based on a final motion vector for a correlated macroblock in a prior frame, where the motion estimator further performs motion estimation for the macroblock based upon the classifying and applies an early termination in accordance with a result of the classifying, wherein the classifying classifies the macroblock of the source frame into one of three potential classes and where the motion estimator outputs a motion vector for the macroblock based on a result of the motion estimation.
20. The video encoder of claim 19 , wherein the video encoder is compliant with at least one of: the H.264 standard, the Moving Picture Experts Group (MPEG)-4, Part 10 standard, or the Advanced Video Coding (AVC) standard.
21. The video encoder of claim 19 , further comprising:
a frame memory for storing the prior frame.
22. The video encoder of claim 21 , further comprising:
a motion compensated predictor coupled to the motion estimator and to the frame memory, for generating a predicted frame in accordance with the prior frame and the motion vector.
23. The video encoder of claim 22 , further comprising:
a subtractor coupled to the input and to the motion compensated predictor for computing a difference between the predicted frame and the source frame;
a transform module coupled to the subtractor for transforming the difference into a set of coefficients;
a quantizer for quantizing the set of coefficients and for providing the set of coefficients, once quantized, to the entropy coder.
24. The video encoder of claim 19 , further comprising:
an entropy coder coupled to the motion estimator for receiving the motion vector and for generating a coded frame from the source frame, in accordance with the motion vector.