A novel binary images edge detection method is presented. It directly clear the inner pixels of the images to obtain the edges. Connectivity-number were sought as detectors, those pixels that have zero-connectivity-number were eliminate as inner pixels from the image. Seventeen well-chosen templates change the computing into efficient exclusive or operations. Experimental results show that it provides less than 2-pixel-wide edges, spends average 8ms on the standard 256-by-256 Lena image. Considered the ratio of details and time consumed, the method shows improved performance than the well known Canny, Roberts and Sobel methods on binary images, and shows potential application on real-time edge detection for binary images.