Abstract: Because of the high similarity between the gradient magnitudes of low-contrast edge and noise, the conventional Canny algorithm for detection, which is based on the double-threshold to detect and connect edges, damages partly the low-contrast edge though it suppresses noise. An improved Canny algorithm is therefore proposed instead of the double-threshold one, which detects and connects the edges in accordance to the difference of the gradient direction between the edge and the noise, so as to protect efficiently the details of all low-contrast edges with the noise suppressed simultaneously. The new approach is thus superior to the conventional one. Simulation tests have verified the effectiveness of the improved Canny algorithm for edge detection.