A novel change detection algorithm using adaptive threshold

摘要

Change detection is a key element in many areas of computer vision that involve multiple views. This paper shows that the use of a fixed correlation threshold for change detection is insufficient to deal with a wider range of scene domains as the best-matched correlation coefficient of an image is contrast dependent. It presents a novel method of adjusting the correlation threshold with respect to the contrast level of the images being analysed in order to enhance the generality of change detection. It establishes the empirical relationship between the contrast level in an image and the best-matched correlation coefficient. It further derives this relationship theoretically, and suggests an efficient method of measuring the image contrast. Experimental results which demonstrate the advantages of the proposed method on real scenes are given.