Research of Similarity Measures for Scene Matching Algorithm

Similarity measure is the most basic and important element of scene matching algorithm. In this paper, the similarity measures for scene matching are summarized. According to the requirements of scene matching system, the performances of mean absolute difference, mean square difference, and normalized product correlation are analyzed through matching simulation experiments. Based on researched of space geometric relationship between the reference image vectror and the real-time image vector, a novel projective similarity measure (Proj) is proposed. Moreover, three scene matching algorithms are designed based on the projective measure. The simulation experimental results demonstrate the good performance of the three algorithms and the validity and efficiency of Proj.