摘要： 针对传统方法难以可靠估计图像中纹理单一像素点视差的问题，提出一种新的基于纹理分析的视差估计算法。与已有方法不同，在以极线约束计算像素点视差时，将极线上纹理单一且近似的像素点合并成直线段，根据连续性和唯一性约束对直线段进行整体匹配，采用直线段的视差得到纹理单一区域的稠密视差图。利用直线段进行整体匹配，提高了比较基元包含的信息量，减少了扫描范围，从而降低误匹配产生的概率和算法时间复杂度。实验结果表明，该方法能提高纹理单一区域稠密视差图的精度，匹配速度快，具有实用价值。

关键词: 视差估计, 纹理单一, 直线段整体匹配

Abstract: According to the problem that it is difficult for traditional algorithms to get disparity of less textured pixels, a novel stereo matching algorithm based on texture analysis is proposed. Different from the existing approaches, less textured and similar pixels are integrated into line segments along a polar line. Line segments are matched based on smooth and unique constraint. The disparity of less textured area is obtained after the disparity of line segments is calculated. The novel algorithm increases the comparative information and decreases the searching range, so that it is able to reduce probability of matching errors and time complexity of algorithm. The Experimental results indicate that the proposed algorithm can improve the matching speed and the disparity accuracy of less textured area, and it is practical.

Key words: disparity estimation, less textured, matching based on line segment

中图分类号: 

• TP391