We propose a method for high-speed image matching, a small number of pixels representing a statistical subset of a template image are used in this method. Generally, when a small number of template pixels are taken into consideration, high-speed matching can be achieved. However, it is often difficult to eliminate the trade-off between speed and reliability in image matching. In order to achieve reliable matching, unique and significant pixels with specific locations and intensities should be selected. For this purpose, we analyze a co-occurrence histogram of multiple local pixels because it provides beneficial information about the probability of simultaneous occurrence. Using the proposed method, we preferentially select pixels with low co-occurrence probability as appropriate template pixels. We also propose a method for determining the approximate probability of co-occurrence of pixels by using some two-dimensional co-occurrence histograms; considerable memory space can be saved when these two-dimensional histograms are used. On the basis of the results of experiments conducted using more than 480 test images, it has been proved that approximately 0.4∼1% of template pixels selected by the proposed method are required to achieve a reasonable object-detection success rate.
J-STAGEがリニューアルされました！https://www.jstage.jst.go.jp/browse/-char/ja/
