Stereo matching between images of large slope based on multi-scale directional wavelet transform

This paper presents a method based on the directional wavelet transform for feature extracting and stereo matching. The original images are transformed by the directional wavelet at two scales, in which the high frequent images in 4 directions may be produced. Using the operator of feature points at three-scales, the feature points in stereo images are extracted. The image matching is at two scales. Firstly the coarse matching was done at scale one. The approximate parallax of both images was obtained. Secondly, at scale zero the feature matching was done based on the describing vectors of 64 parameters. Then mismatched points were eliminated in the epipolar restriction. The experiments in airplane, unpiloted aircraft and close-rang stereo images show that the approach is accurate and robust for extracting keypoints and stereo matching. The matching results is rotation or scaling invariant.