Robust object detection and segmentation by peripheral increment sign correlation image

This paper proposes the peripheral increment sign correlation image, which can be used to evaluate sign changes of the neighborhood brightness. Based on the technique, a robust method is presented for detection and separation of the object emerging from the image time sequence, with the following features. (1) The result is not affected by the brightness distribution of the emerging object. (2) The result is not affected by brightness change of the background. The peripheral increment sign correlation image is constructed from only the trend of the brightness changes in the neighborhood of the pixel under consideration. Consequently, the image is highly robust to brightness changes over the sequence, and the similarity of the original texture pattern can be detected even if there is a brightness change. Furthermore, it does not require complex preprocessing or parameter setting, and a high-density difference image can be directly derived. Through experiments using real images with various conditions, it is verified that difference extraction which is robust to brightness change can be realized, indicating the effectiveness of the approach. An application experiment is performed to extract the emerging object area in a scene, and it is verified that segmentation can be performed while retaining good contours and continuity by adding a relatively simple filter sequence as postprocessing. © 2004 Wiley Periodicals, Inc.