We propose a high-speed template matching method using small number of pixels that represent statistical subset of an original template image. Generally, to reduce the number of template pixels means low computational cost of matching. However, high-speed and high-reliability often have trade-off relation in actual situations. In order to realize reliable matching, it is important to extract few pixels that have unique characteristics about their location and intensity. For this purpose, analysis of co-occurrence histogram for local combination of multiple pixels is useful, because it provides beneficial information about simultaneous occurrence probability. In the proposed method, pixels with low co-occurrence probability are preferentially extracted as significant template pixels used for matching process. Also we propose a method to approximate n-pixels co-occurrence probability using some two-dimensional co-occurrence histograms to save memory space. Through some experiments using more than 480 test images, it has been proved that approximately 0.2 to 1% of template pixels extracted by proposed method can achieve practical performance. The recognition success rate is 96.6%, and the processing time is 15 msec (by Core 2 Duo 3.16 GHz).
Extraction of unique pixels based on co-occurrence probability for high-speed template matching
01.10.2010
2554568 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Recognition by matching dense, oriented edge pixels
IEEE | 1995
|Recognition by Matching Dense, Oriented Edge Pixels
British Library Conference Proceedings | 1995
|Pixels Classification for Moving Object Extraction
IEEE | 2005
|