多通道置信传播算法的核线影像密集匹配  

Dense stereo matching of epipolar images by multichannel belief propagation algorithm

作　　者：张漫[1] 胡腾[2] 

机构地区：[1]北京信息职业技术学院,北京100015 [2]中国地质大学(北京),北京100083

出　　处：《测绘科学》2017年第6期55-61,111,共8页Science of Surveying and Mapping

基　　金：国家自然科学基金青年基金项目(41602215);测绘遥感信息工程国家重点实验室2014开放研究基金项目((13)重03)

摘　　要：针对航空核线影像的稠密匹配问题,该文提出了一种基于控制点与RGB三色置信传播模型的多通道置信传播算法。该算法利用点特征匹配算子提取影像的部分同名点作为区域控制点,随后将RGB 3个颜色分量都作为置信传播模型的独立基础变量,同时增加邻域像素点对的颜色差异的贡献。两组核线影像匹配实验结果表明,所提算法在使用较少参数的情况下,降低了同名点存在颜色差异造成的干扰,利用置信传播原理提高了同名点的匹配正确率,能有效进行航空核线影像的逐像素密集匹配。Aiming at the problem of dense stereo matching of aerial epipolar images, a multichannel belief propagation （BP）algorithm based on control points and RGB components propagation model was proposed in this paper. Parts of corresponding points generated by feature detection operator were used as control points. Three color components R, G and B were used as independent and basic variables of belief propagation model. The contribution of neighboring pixels was increased to depress the undesirable interference from the corresponding pixels＇ color differences. Experimental results of two groups of epipolar images matching showed that multichannel BP algorithm is efficient for dense stereo matching of aerial epipolar images; under the condition of few parameters, the proposed algorithm could reduce the direct impact of color differences between two matching pixels and improve the matching correctness.

关 键 词：稠密立体匹配 贝叶斯模型 控制点 多通道 置信传播 

分 类 号：P751[交通运输工程—港口、海岸及近海工程]