Abstract: Edge detection is used in image processing and computer vision. The typical Sobel edge detection algorithm in the digital image processing was analyzed. An improved algorithm for the image edge detection based on Sobel algorithm was proposed to overcome the disadvantages, which the obtained edges of the image is thick and sensitive to the noise. The edge types that exist in the real images are described with mathematical models and the models of the continuous edges are regarded as a research subject. A template for detecting the direction of the image edges was rebuilt. The thinning processing for the gradient histograms of images was adopted, to improve the low accuracy of locating the edge position, caused by the traditional Sobel edge detection which was based on oneorder derivative′s maximum value or two-order derivative′s zero-crossing. The simulated result shows that the algorithm has a better immunity to the noise jamming of images and the edge position extracted by the algorithm is accurate.