一种新的用于立体图像编码的视差估计算法

A Novel Disparity Estimation Algorithm for Stereo Image Encoding

视差估计是立体视觉和立体图像编码领域的研究重点 ,其估计的准确程度直接影响编码的精度和效率 .该文在研究传统估计算法的基础上 ,提出了一种新的带极线约束的DT基立体图像视差估计算法 .该算法首先把立体图像中的左图像作为参考图像 ,用DT网格进行描述 .然后 ,利用极线约束条件在右图像中估计出各三角形顶点的对应点 ,整个估计过程要求DT结构保持不变 .通过估计出来的对应点计算出这些顶点的视差矢量 ,再用六参数仿射变换估计出三角形内部各点的视差矢量 .编码时 ,只需编码各顶点的视差矢量 .从实验结果看 ,这种新算法同传统的“块”匹配视差估计算法相比 ,视差矢量的估计更加准确 ,视差图中主要物体的轮廓更加清晰 ,有利于进一步的物体分割 .

Disparity estimation is important in the stereo vision and stereo image coding whose results influence the precision and efficiency of the coding system. This paper introduces a novel Epipolar line-Constraint DT-based Stereo Image Disparity Estimation Algorithm after studying the traditional disparity estimation algorithms. This algorithm uses the DT mesh and epipolar line constraint between two corresponding points. The first step of this algorithm is to describe the left image by DT mesh, which is regarded as the reference image. Secondly, the corresponding vertexes of triangles are estimated by using the theory of epipolar line-constraint within the right image. In the course of estimation, the DT structure must keep changeless. Finally, the disparity vectors of vertexes are calculated and the disparity vectors inner the triangles are estimated by six-parameters affine transformation. While encoding, only the disparity vectors of vertexes are coded, which carry compression’s point. Some real stereo images are used in the experiment, and the results show that the estimation veracity of this algorithm is higher than the traditional algorithms, which are based on “block” matching, and the object’s contour in the disparity map is clearer than before, which means that the disparity map produced by this algorithm is more beneficial for object’s segmentation. As a result, this algorithm is adaptive to coding system.