A new mean filter ratio technique for edge detection and foreground extraction

Edge detection is the primary step in image segmentation and target detection applications. The edge operators proposed so far in the literature, namely, Canny, Sobel, Prewitt, provide a number of unwanted edges which complicate the foreground object detection process. In this paper, a novel technique is proposed for edge detection and foreground segmentation employing two mean filters of different window sizes. A ratio of the filtered images is taken and normalized. Then a threshold is applied on the histogram of the resultant image to derive the final output which can detect the edges and hence separate the foreground from the background. Performance of the proposed method has been investigated through computer simulation and compared with other existing edge detection techniques using complex reallife image sequences, which verifies that the technique provides better detection results for any input scene.