The Application Of Image Feature Extraction And Segmentation Algorithm In Apple Image

Our country is a big agricultural country, also is a big populous country in the mean time, the agriculture was related to political stable and economic prosperous of our country. Along with the growth of request of the high quality and safe agricultural product, the ability to accurately and fast authenticate the quality of agricultural products keeps increasing. In recent years, the research of agricultural products detecting based of computer sense of vision technical is very active. The key technique which makes use of a machine sense of vision's technique to carry on agricultural product external appearance an examination is to get the valid agricultural product images' information, and adopt a suitable method of image segmentation. In order to accurately extract information of agricultural product, we have to use a valid feature extraction method and a valid image segmentation method.The color and texture feature of apple image are analyzed in this thesis. A method of feature extraction means based on gray and texture is proposed aiming at characteristic of apple image, clustered with Fuzzy Adaptive Resonance Theory. Extraction of apple image is finished after image smoothing. The main contributions include:Review the development of image processing researches, summarize, sort out image feature extraction, image segmentation and image smoothing. And the image segmentation methods are classified introduce: on the base of threshold, on the base of Otsu, on the base of edge detection, on the base of region and on the base of nerve network.It is pointed out the defect of image feature extraction on the base of color. Anglicizing apple image, extracting gray value, which is the level of light, and energy, which is the weight of distribution of apple image. And deal with the combined feature with complement coding.Adaptive Resonance Theory and Fuzzy Adaptive Resonance Theory are introduced. The construction of network and arithmetic are studied in detail. Training the Fuzzy ART nerve network with fuzzy feature vector, comprised of gray and texture based on gray level co-occurrence matrix and finish the first segmentation. Then process the first segmentation result with binaryzation and median filter, getting rid of noise. In this way, we can get the defect of apple image. Put the system of apple image defect segmentation into practice, analysis the parameters of the system and compare with some classic image segmentation arithmetic. Then the results of the experiments are analyzed.