A Study of Disparity Map Based on Improved Dynamic Programming Algorithm

摘要

In recent years, as a related technology, stereo display(3D) has been growing more and more rapidly with the development of internet. However, some problems have arisen, either. To solve the problem about holes and poor matching in the process of virtual viewpoint generation, a new method for improving the disparity map of Dynamic Programming algorithm (DP) is proposed. Firstly, the RGB images of the left and right viewpoints are color-divided in the Hue-Intensity-Saturation space (HIS), and then the images is stereo-matched by the proposed method that DP algorithm and the Sum of absolute differences algorithm (SAD) to generate an initial disparity map. The cost function based on HIS space is added to the calculation process of matching cost, which increases the precision of disparity calculation. Then the linear interpolation and median filtering are used for eliminating the holes in the disparity map. The experimental simulation shows that compared with the traditional algorithm, the improved algorithm optimizes the quality of the synthetic disparity map, and reduces the mismatching rate. In addition to these, compared with standard disparity map, the improved method preserves the information of the initial image to some extent.