融合组合代价与树形聚合的鲁棒半全局匹配算法OA

This paper addressed the limitations of traditional semi-global matching algorithms that rely on a single matching cost,particularly in complex scenes where high texture similarity,strong sensitivity to lighting,and significant radiometric differences can easily cause matching discrepancies.An improved approach was proposed,which constructed a composite matching cost function that integrates pixel grayscale sum of squares and mutual information to evaluate image similarity using dual criteria.Additionally,a tree-based cost aggregation method was introduced,utilizing a minimum spanning tree to search for the globally optimal path.Tests on standard datasets such as the Middlebury dataset demonstrated the significant advantages of the improved algorithm:compared to traditional single cost measures,the combined cost model reduced the average mismatch rate in areas with repetitive textures and improved the disparity estimation accuracy in scenarios with strong lighting changes.Experimental data confirmed that this optimization method,which merges multiple feature criteria with a tree-based aggregation strategy,effectively solves the adaptability problem of traditional single measures in complex imaging environments,providing a new technological path for stereo matching.