The Study Of Color Image Edge Detection Based On Comentropy And Neural Network

Edge detection is one of the most important and difficult tasks in computer vision. Itrequires accurate edge detection and classification. Image information is tremendous and theedge information is a tight description and also is the basic characteristics of image, it containsuseful information for cognition. According to the theoretical analysis and experimental results,the existing color image detection algorithms easily lose the low-contrast edges and underutilizethe color information. This thesis is targeted for an appropriate solution to these problems.The essence of color image edge detection is the definition of color difference andcalculation. This paper discusses the utilization of information entropy and the neural networkfor color image edge detection. Because the size and direction of color difference is difficult todefine, the method used to describe image edge features based on eigenvectors and regionalcoherence measure is proposed. As for the problem of how to use obtained data to apply tomeasure the edge detection, color image edge detection method based on neural network isproposed. The specific content as follows:This paper reviews the current typical edge detection algorithm, and focuses on the colorimage detection method, introduces color space, neural network and other basic knowledge.Based on HIS and RGB color space, the method of the utilization of each color component indifferent color space to measure the amount of information is proposed, for different color space,calculated the corresponding color component, through the introduction of fuzzy entropy, andconstructed a set of information measure based on fuzzy entropy weight to describe the imagefeatures quantitatively. At the same time the method is applied to describe the colorcharacteristics successfully. This method used the various color components of each color spacecompletely. In particular, it is reasonable to use the correlation between color components, ratherthan the proportional use of each color component.Based on the obtained information measure that can be described the image edge, BP neuralnetwork is used for edge detecting. Firstly, BP neural network is trained with some eigenvectorof the four component vectors, and then the trained BP neural network is used for edge detectiondirectly.Performance evaluation of the merits of edge detection can be divided into subjectiveevaluation and objective evaluation. Paper used subjective evaluation and objective evaluationmethods to evaluate the new color image edge detection algorithm. Using the subjectiveevaluation of the observing of obtained images after treatment, the experimental results showthat this method can obtain more complete outline of the image to retain the original color thanother algorithms. Using objective evaluation method by edge connectivity and anti-noise performance analysis, it found that the new method significantly better than the other edgedetection algorithm.This fuzzy entropy and BP neural network were used for color image edge detection in this paper, goodresults were obtained, and this algorithm enriched the color image edge detection method. The next step willbe to explore the possible outcomes of color image edge detection based on other information entropy andother neural network and analyze the function of all kinds of information entropy in different neural networks,to obtain the most suitable information for edge detection in different types of color images, and to reduce theblindness and randomness in the choice of algorithm of color image edge detection.