Coarse-to-fine Scheme

 

It has been shown that a multi-resolution or pyramid data structure approach to stereo matching is faster than one without multi-resolution [13], as the search range in each level is small. Besides fast computation, a more accurate disparity map can be obtained by exploiting multi-resolution. The upper levels of the pyramids are ideal to get an overview of the image scene. The details can be found down the pyramid at higher resolution [14].

During the process of projecting the disparity map from the current level of the pyramid to the next (if current level is not level 0), the image size was scaled up by the value of r (reduction ratio), and the disparity value by the same r. The disparity value where the position (i,j) of the new image is not a multiple of r was obtained by linear interpolation.

Our proposed algorithm for stereo matching is:

1. Build pyramids with k levels, and the reduction ratio of r; The upper or coarse resolution levels are obtained by averaging the corresponding r r pixels in the previous level;
2. Initialize the disparity map as zero for level k;
3. Perform image matching using the method described in sections 3.1-3.3;
4. If , propagate the disparity map to the next level using linear interpolation, set k=k-1 and then go back to step 3; otherwise go to step 5;
5. Smooth the disparity map using a median filter.