基于冗余离散小波变换的立体图像视差估计

为提高立体图像视差估计的准确度及效率，从图像特征点入手，结合DT（delaunay triangle）网格技术，提出基于冗余离散小波变换的立体图像视差估计法．对参考图像进行冗余离散小波变换后，根据其平移不变性和各级子带大小相同的特点，通过各级高频子带相乘、叠加，再取相应闽值，得到参考图像特征点，并形成DT网格；以特征点为中心在目标图像中进行视差估计，得到各特征点的视差矢量，再结合仿射变换得到三角形内部各点的视差矢量，从而形成浓密的视差图．实验结果表明，根据视差图重建的目标图像PSNR值要高于传统的SAD块匹配法和Canny特征点法5～10db；对于CIF格式的立体图像，算法在Matlab7．0下时间消耗为15s．该算法在视差估计准确度和运算速度上具有较好性能，可应用于立体图像处理领域

Redundant Discrete Wavelet Transform Based Disparity Estimation for Stereo Images

To improve the accuracy and efficiency of the disparity estimation of stereo images, combined with feature points of the image and DT（delaunay triangle）technology,a method based on redundant discrete wavelet transform （RDWT） was presented. According to the shift invariance and the same size of each sub-band in an RDWT, different levels of horizontal, vertical, and diagonal sub-bands were multiplied and combined, then the feature points were obtained by choosing a threshold and the DT mesh was generated. Disparity vectors of each feature point of the target image could be obtained by feature-matching method, and the disparity vectors inside the triangles were obtained by affine transformation to generate a dense disparity map. Experimental results show that the PSNR value of reconstructed image according to the disparity map is 5-10 db higher than the traditional SAD block-matching method and Canny feature-point method. For a pair of stereo images in the CIF form, the run-time of disparity estimation is 15 s in Matlab 7.0. Therefore the method works well in the aspects of disparity estimation accuracy and operation speed, which can be used in the stereo image processing.