A Robust Matching Method for Low-Textured Images Based on Co-Occurrence Probability of Geometry-Optimized Pixel Patterns

In this paper, we propose a template matching algorithm that is applicable to low-texture images, such as range images. Among high-speed template matching methods, the co-occurrence probability-based template matching (CPTM) method is useful and effective. This method uses some sets of selected pixel patterns that have relatively low occurrence probability in a template image. By using such a small number of distinctive data, reliable matching has been achieved, in addition to high-speed processing. However, this method is problematic insofar as the extraction of distinctive pixels is difficult when the distribution of the occurrence probability is uniform. For example, this is frequently found in range images. We improve the CPTM method in order to address this problem. A key idea is to optimize geometric pixel relations in the pixel patterns when the proposed method calculates the occurrence probability of pixel patterns. Experimental results have confirmed that the proposed method increases the detection rate from 73% to 90% without sacrificing its capacity for high speed. This shows that the performance of our method is better than that of conventional methods.