The two thresholds of classical Canny operator need to be set manually, which limits the application of this algorithm. Therefore, many researches about how to choose threshold adaptively are done to solve this problem. Based on the gradient histogram, a method of threshold-adaptable edge detection is proposed. This method is on the basis of gradient histogram difference diagram with adaptive image classification techniques. It not only automatically sets the two thresholds, but also avoids disconnected or false edges in detection. Experiments prove that the method is threshold-adaptive and advantageous for edge detection in color image whose pixels of larger gradient amplitude are mainly located in the edge between the target and background.