New approach to object recognition in textured images

This paper describes an approach to object recognition in textured images. The method is based on image matching via distance transform. The proposed matching scheme is an extension and combination of the conventional methods in image processing. Interesting points are detected to replace edge pixels as image feature pixels in distance transform for the matching measurement. A mask based stochastic method is introduced to extract texture features. The detection of interesting points for matching is then performed on such a texture feature image named texture energy image. A dynamic thresholding procedure is further applied to guide the matching. Our experimental results demonstrate that the combination of texture feature extraction and interesting points detection provides a better solution to the search of the best matching between two textured images. In addition, such an algorithm is simple to implement and quite insensitive to noise and other disturbances.