Prediction of the suitability for image-matching based on self-similarity of vision contents

摘要

This paper is concerned with predicting the suitability for image matching to measure the extent to which a specific area in the input image is difficult for image matching. First, we describe a new reinforced anti-noise image matching technique, which is composed of Vision Content Description for Matching (VCDM) and Intuition Level-wise Critical Matching (ILCM), because we also must perform the image matching in order to predict it's suitability. Next, we construct a prediction model for measuring the suitability for image-matching based on local and global self-similarities of vision contents and the analysis of mismatching elements. Prediction results are represented in two different formats: bitmap and contour map. From various experiments on block template matching, we found that (1) our proposed image matching technique results in both a high degree of accuracy and robustness against noise and (2) the suitability measure for image matching is consistent with the valuable contents in the input image. From these results, it is believed that the predicted suitability measure can successfully serve as a guide for a further process in object matching.