【摘要】 本文将近年来兴起的小波理论应用于双目立体视觉技术，提出了一种新立体匹配算法，并将其应用于金属材料断口ＳＥＭ图像体视匹配，获得了较好的效果。该算法以二进小波变换局部模极大值边缘点作为匹配基元，重要的是可以从变换系数中推出水平和垂直方向分量的幅度变化信息，合称为边缘分量幅度变化相似约束，他们直接反映了图像结构的变化，对于复杂景物的匹配尤其重要。算法综合利用边缘点的特征属性和一系列约束条件，从边缘点局部相似和全局匹配相容两个方面解决模糊匹配问题。实验结果表明，该算法稳定可靠，具有较强的处理复杂景物的能力更多还原

【Abstract】 This paper proposes a new algorithm for stereo matching which utilizes the wavelet transform theory developed recently. The multiscales local modulus maximums of 2 D dyadic wavelet transform are used as matching primitives. More importantly, from the two components of the wavelet transform, we get two useful constraints called "local amplitude energy constraints", which can greatly remove those potential matches that would cause mismatches. The matching approach utilized the characters of wavelet transform local modulus maximums and the power matching constraints to resolve the ambiguity of matching in local feature similar and global matching consistency. Experimental results show that the algorithm is simple, reliable and capable of adaptable coping with complicated real scenes.更多还原