A Quick stereo Corresponding Method Based On Lifting Wavelet Schema

摘要

We propose an improved method on Stereo Corresponding, which is based on Lifting Wavelet Schema. Stereo correspondence has long been an interesting topic in computer vision. A lot of corresponding algorithms have been suggested on how to estimate it. Yet among these algorithms, much focuses have been putted on how to make stereo matching more accurate, while attracted less attention on how to improve efficiency. We propose a fast and robust method on stereo correspondence while computing complexity is reduced greatly. Thanks to Lifting Wavelet Schema, we can scales the input images to a series of subspace with different resolution. One advantages of wavelet translation is that the image can be reconstructed perfectly. Wavelet translation can also remove the noise signals in images due to various unknown facts. Graph Cut is a good solution on labeling schema; it is perfect for computing disparities between two images rectified. With these two techniques, we first scale the images, and then corresponding method based on graph cut is applied on these images. Because subspace images lose information of disparity, after an estimated disparity image is on hand, we'd tune the estimated disparity to get a more really disparity image. Then we can get a satisfied disparity image while less time is elapsed. Experiments show that our method is robust and efficiency.