Stereo Matching Based On Phase Information

3D video system has recently emerged as a way to enhance the watching experience, which is becoming a focus of video applications. And stereo matching is one of the most active research areas in 3D video system, a very important way to obtain the depth information of the scene in the stereo images. Local phase could not only reflect the original image structure completely, but also resist the geometry and noises distortion. Therefore stereo matching approach based on phases is a key research direction. In this thesis, we investigated the main algorithms and presented some improved solutions on phase-based stereo matching.Phase-based matching method is developed in recent two decades, which is based on Fourier shift invariance theorem. By using complex band-pass filters, we can get the phase information in different frequency bands. In the first part of this thesis, we choose Dual-Tree Complex Wavelet Transform (DT CWT), which is approximate shift invariance, to extract phases. A coarse-to-fine matching solution based on phase of different resolutions is presented, effectively inhibited the most two common problems in phase-based matching——singularities and phase wrap. We call it"BestMatch approach". The results of this solution show the general disparity map between the stereo images. However, for the disparity edge areas and texture lacking areas, some improvements should be done to amend the bad estimated pixels.As we mentioned, there are plenty of matching mistakes at edge areas and texture-lacking areas by just using phase information. To overcome this disadvantage, we introduced color segmentation and plane fitting, on the assumption that the scene structure can be approximated by a set of non-overlapping planes in the disparity space and that each plane is coincident with at least one homogeneous color segment in the reference image. Here, we present a whole stereo matching scheme, which includes two parts: phase-based matching and initial disparity amending. This new approach shows higher accuracy than most of the other recent phase-based algorithms, and the disparity at edge areas could be effectively shaped and clearly shown in the map.At the last part of the thesis, we introduced the conceptions of quaternion and quaternion wavelets. And we constructed a kind of quaternion wavelets by using the tensor product of DT CWT to derive 2 dimensional disparity maps based on quaternion phases. A 2D sparse disparity estimation approach is proposed for uncalibrated stereo images in this part. The experimental results show that this method has accurate estimation and could play a role in free view disparity estimation as well as motion estimation.