Study On The Segmentation And Recognition Method Of Color Blood Cell Image

With the rapid development of computer relative technology in recent years, image processing technology are more and more mature, a new era of digital image processing has arrived, these new technology also penetrate into the medical area. Medical image processing is a new technique arising in the past 30 years, and the blood cell automatic recognition using computer to process medical micrograph is a representative topic. Integrated with computer image processing, virtual realization and computer network technology, use computer to analyze, process medical image is also a hot spot. This thesis discussed how to realize the automatic analysis of 24-bit true color blood cell micrograph using computer image processing and pattern recognition techniques. The main content consists of:①The image preprocess method related to blood cell micrograph is discussed, including grey scale stretch, mean filter, math morphologic algorithm such as opening and closing. Mean filter is applied to the image in horizontal and vertical direction, and large scale is used to remove the affection of noise.②Based on the preprocessed image, several segmentation algorithm is discussed such as threshold segmentation, seed point region growing algorithm, contour trace, edge detection and so on. Especially, the threshold limit and local maximum method are used to detect the edge in two directions respectively. Connect the edges detected in two directions to take full advantage of its concavity and convexity and edge type, to segment cell effectively.③The character description for the blood cell segmented image is discussed, and several features being able to reflect the distinction of blood cell at conformation, color and texture are extracted.④The image recogniton approach is discussed such as the decision-tree classification algorithm of the image recognition theory, support vector machines, artificial neural network recognition algorithms. Using the Matlab NN tool box, experiment is given for feed-forward BP network, several kinds of learning algorithms are used to training the network, meanwhile, the results are compared with each others, the experiment results show that the L-M algorithms has fast convergence speed and better classification accuracy.