Method and apparatus for estimating disparity

1. A disparity estimation method performed by a processor, the disparity estimation method comprising:

extracting a first image patch comprising a reference pixel from a first image;

extracting a second image patch comprising a target pixel corresponding to the reference pixel from a second image;

estimating, in an integer-pixel unit, an initial disparity between the reference pixel and the target pixel; and

estimating, in a sub-pixel unit smaller than the integer-pixel unit of the initial disparity, a residual of the initial disparity based on the first image patch and the second image patch using a residual model, the residual being an estimated difference between the initial disparity and an actual disparity between the reference pixel and the target pixel.

2. The disparity estimation method of claim 1 , further comprising determining a final disparity between the reference pixel and the target pixel by correcting the initial disparity based on the estimated residual.

3. The disparity estimation method of claim 2 , further comprising calculating a depth corresponding to each of pixels of the first image and the second image based on the final disparity.

4. The disparity estimation method of claim 1 , further comprising estimating the initial disparity between the reference pixel of the first image and the target pixel of the second image.

5. The disparity estimation method of claim 4 , wherein the estimating of the initial disparity comprises:

determining a search range in the second image;

comparing a reference image patch comprising the reference pixel to each of candidate image patches respectively corresponding to pixels included in the search range; and

determining the target pixel in the second image based on a result of the comparing.

6. The disparity estimation method of claim 1 , wherein the estimating of the residual comprises:

extracting feature data from the first image patch and the second image patch using a feature model; and

estimating the residual from the feature data using the residual model.

7. The disparity estimation method of claim 1 , further comprising estimating false disparity information of the initial disparity from the first image patch and the second image patch using a false disparity detection model.

8. The disparity estimation method of claim 7 , further comprising excluding the initial disparity from further use in response to determining that the initial disparity is false based on the estimated false disparity information.

9. A disparity estimation method performed by a processor, the disparity estimation method comprising:

extracting a first image patch comprising a reference pixel from a first image;

extracting a second image patch comprising a target pixel corresponding to the reference pixel from a second image;

estimating a residual of an initial disparity between the reference pixel and the target pixel from the first image patch and the second image patch using a residual model, the residual being an estimated difference between the initial disparity and an actual disparity between the reference pixel and the target pixel; and

estimating the initial disparity in an integer-pixel unit,

wherein the estimating of the residual comprises estimating the residual in a sub-pixel unit.

10. The disparity estimation method of claim 1 , wherein the extracting of the first image patch comprises:

extracting a feature point from the first image; and

determining a pixel corresponding to the feature point in the first image as the reference pixel.

11. A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause the processor to perform the disparity estimation method of claim 1 .

12. A disparity estimation apparatus comprising:

a processor configured to:

extract a first image patch comprising a reference pixel from a first image,

extract a second image patch comprising a target pixel corresponding to the reference pixel from a second image,

estimate, in an integer-pixel unit, an initial disparity between the reference pixel and the target pixel, and

estimate, in a sub-pixel unit smaller than the integer-pixel unit of the initial disparity, a residual of the initial disparity based on the first image patch and the second image patch using a residual model, the residual being an estimated difference between the initial disparity and an actual disparity between the reference pixel and the target pixel.

13. A disparity estimation method performed by a processor, the disparity estimation method comprising:

estimating an initial disparity in a single-pixel unit between a reference pixel in a first image and a target pixel in a second image, the target pixel corresponding to the reference pixel;

estimating a residual in a sub-pixel unit smaller than the single-pixel unit from the first image and the second image using a residual model, the residual being an estimated difference between the initial disparity and an actual disparity between the reference pixel and the reference pixel; and

obtaining another disparity by combining the initial disparity with the residual.

14. The disparity estimation method of claim 13 , wherein the estimating of the residual comprises:

extracting a first image patch comprising the reference pixel from the first image;

extracting a second image patch comprising the target pixel from the second image; and

estimating the residual from the first image patch and the second image patch using the residual model.

15. The disparity estimation method of claim 13 , further comprising:

estimating the false disparity information indicating a probability that the initial disparity is false from the first image and the second image;

excluding the initial disparity from further use in response to the false disparity information exceeding a false threshold level; and

estimating the residual in response to the false disparity information not exceeding the false threshold level.

16. The disparity estimation method of claim 13 , further comprising:

estimating false disparity information indicating a probability that the initial disparity is false from the first image and the second image;

excluding the initial disparity from further use in response to the false disparity information exceeding a false threshold level; and

estimating the residual in response to the false disparity information not exceeding the false threshold level.

17. The disparity estimation apparatus of claim 12 , further comprising one or more image sensors configured to acquire the first image and the second image.

18. The disparity estimation method of claim 1 , wherein the extracting of the second image patch comprises extracting the second image patch based on the initial disparity.

19. The disparity estimation apparatus of claim 17 , wherein the processor is further configured to determine a final disparity between the reference pixel and the target pixel by correcting the initial disparity based on the estimated residual.

20. The disparity estimation apparatus of claim 19 , wherein the processor is further configured to calculate a depth corresponding to each of pixels of the first image and the second image based on the final disparity.

21. The disparity estimation apparatus of claim 17 , wherein the processor is further configured to estimate the initial disparity between the reference pixel of the first image and the target pixel of the second image.

22. The disparity estimation apparatus of claim 21 , wherein the processor is further configured to:

determine a search range in the second image,

compare a reference image patch comprising the reference pixel to each of candidate image patches respectively corresponding to pixels included in the search range, and

determine the target pixel in the second image based on a result of the comparing.

23. The disparity estimation apparatus of claim 17 , wherein the processor is further configured to:

extract feature data from the first image patch and the second image patch using a feature model, and

estimate the residual from the feature data using the residual model.

24. The disparity estimation apparatus of claim 17 , wherein the processor is further configured to estimate false disparity information of the initial disparity from the first image patch and the second image patch using a false disparity detection model.

25. The disparity estimation apparatus of claim 24 , wherein the processor is further configured to exclude the initial disparity from further use in response to determining that the initial disparity is false based on the estimated false disparity information.