Research On Image Edge Detection Based On Grey Correlation Degree

Image edge detection which belongs to the category of feature extraction is an important research field of digital image processing, in addition, it is also an important foundation of computer vision. In practical application, image edge which contains the primary information of image has a great significance.Meanwhile, the image edge detection is also a puzzle in digital image processing. The achievement of obtaining the ideal edge effectively is the primary problem for edge detection algorithm.The spatial differential is used to calculate convolution in the principle of classical edge detection algorithm. The gray mutation which constitute the gray edge is an important characteristic of image edge.The classic gray edge operators use derivative operator to detect the edge of gray change. With the development of technology, a great many of new edge detection theory has been proposed, such as edge detection algorithm based on fractal geometry, edge detection algorithm based on Wavelet transform and wavelet packet, edge detection algorithm based on morphological, However, these methods also have their limitations respectively.The grey system is an incomplete information model, and the grey relational analysis with characteristics of "little data, poor information", is used to analyze the grey system with mathematical theory. Through referencing to the known information, the development trend of unknown information is obtained quantitatively.In this paper, the grey system theory and the improved traditional edge operator are combined, as well as the MLP neural network. And the main work is as follows:(1) Aiming at the classical edge detection operators for edge information is rich in image and weak edge image detection effect is not ideal. Firstly, this paper combined with the grey relational theory develops, extend the traditional Sobel template from 0° degrees and 90° to 0°, 22.5°, 45°, 67.5°,90°, 112.5°, 135°, 157.5° 180°, 202.5°, 225°, 247.5°, 270°, 292.5°, 315°, 337.5°, in order to increase the direction information. Secondly, taking the extended template as the reference sequence of grey relational analysis theory, and calculate the max absolute correlation degree, Then, set the appropriate threshold and the image edge is obtained.(2) To the problem of computation cost and time consumption of the algorithm combined the multi-direction Sobel operator with the grey correlation theory. The image obtained through combing the method of multi-directional Sobel operator and the grey correlation theory is considered as the training set sample of MLP neural network in this paper. By mounts of training, the reasonable parameters of the MLP neural network is set. Finally, with the help of trained neural network the image is detected. The detection time is reduced effectively and the better results is achieved.