The Improved Differential Operator Fuse With Morphology Edge Detection Method

Edge is the most basic feature of the image, containing most information of the image, thus it is the important basis of image segmentation and image analysis. Edge detection has become a priority for the image analysis and recognition, has very important application in image processing. But edge detection is difficult to filter out noise and meanwhile enhancing edges in edge detection. This paper put forward the improved algorithm for the shortcoming of traditional mathematical morphology and differential operator, and then fused two kinds of improved algorithm. Good results have been achieved. Specific work is as follows.Differential operator for edge detection exists different defects. Taking Canny operator as an example, there are two drawbacks. It can not decimate the effect of part noise, and edge details will be lost at the same time when the pseudo edge is detected. It is difficult to set Gaussian filter parameters, and it lack adaptability for different images. Aimed at the defects of the traditional Canny edge detection, this paper puts forward an improved Canny edge detection algorithm. First, this paper gives a new method on adaptive image block based on threshold value. Next, by proposing a new hybrid filtering about adaptive median and morphology, we adopt this hybrid filtering to smooth the image with noise. Then, we add the gradient information of two oblique directions, so that the gradient information is completer. Last, by using the double threshold value to process the gradient image which is after non-maxima suppression, we obtain the final edge. For the image with noise, this improved algorithm not only can filter out noise well, but also the image edge is continuous, smooth, clear.In morphology, considering that the elements with different structures and scales play different roles in removing noise and retaining image details, an adaptive algorithm of edge detection based on multiple structures and multi-scale element was presented. Firstly, improve the existing edge detection operator. Secondly, the weight coefficient of different scales and different shapes were determined by using the morphological difference method and calculating the edge information entropy after detecting. Finally, the edge can get by fusing detected edges. For the image with noise, the algorithm can filter out noise effectively, and the objective evaluation and visual effect are good.Finally, the Canny algorithm given above has better performance in objective evaluation, such as standard deviation and peak signal-to-noise ratio, etc. The improved morphology algorithm performs better in average gradient, the degree of distortion and the correlation coefficient. To improve the accuracy of edge detection and enrich the image edge, this paper tried to fuse these two improved algorithms to make the edge image in terms of objective evaluation(information entropy, the degree of distortion and the correlation coefficient, etc.) better than the former two methods, and satisfied in visual effect.