The Research Of Color Image Segmentation Based On BLR-DRF Model

At present, with the rapid development of computer science and technologys, graphic and image processing technology has applied to almost all the fields of human activities, this technology gets the most widely use and development, especially in the fields of medical symptoms diagnostics, the computer automatic control, biological engineering and the pattern recognition, etc.As a key link in the image processing system, image segmentation has been highly valued for many years. Meanwhile, with the development of science and technology, the color images use frequently in some fields, the results of color image segmentation has more value in use, due to the more informations contained in color images than in the gray level images, thus, the color image segmentation has drawn the increasing attention.There are many classical algorithms that used in the gray level image had been put forward to resolve the segmenting problem, including the edge-based methods and region-based methods, etc. The conventional segmentation algorithms have the incomparable advantages in solving some specific problems, but there are also corresponding disadvantages such as the edge-based methods can not form the close edges, the histogram threshold method based on region segmentation depends on the selection of the threshold value. The regional growth methods depend on the selection of the seeds and the judge of the search direction, etc. These defects restrict the application of these algorithms in the actual corresponding fields. Since the problem of image segmentation is implied uncertainty, it attracted many researchers to create different image models and algorithms for a variety of applications. The new trend is introducing new approachs and new concepts into the image segmentation, and optimal combination of various methods to achieve the desired segmentation result.Based on the applying section of this paper, the color segmentation system of pasture plant diseases are obtained to match the requirement of the pasture plant disease monitoring system, which is one part of the national 863 project in the key technology research of resource-saving management of the digitized northeast grassland. According to needs of the disease segmentation system of the project, we use the statistical and pattern recognition theory, with the theoretical basis of the logistic regression and the Markov Random Fields,Bounded Logistic Regression and Discriminative Random Field is the improved theory, to combine the BLR model and the DRF model, by solving the pasture plant Leaf Disease Spots Segmentation as a constructed on BLR-DRF model. Obtained by experiment,the algorithm is robust to noises and can adapt to the changing environment.The main works as follows:(1)we studied the common methods of image segmentation at home and abroad, by reading a lot of literature, to obtain a brief introduction to the currently available methods for color image segmentation including edge-based image segmentation method, region-based image segmentation and a variety of classification methods based on the combination of specific theories such as morphology, wavelet transform, genetic algorithms, clustering algorithms, neural networks, fuzzy mathematics and artificial intelligence and other areas of research.(2) By studying the basic theory and model of logistic regression, the improved bounded logistic regression method is introduced to the color image segmentation system. According to the theory of statistical decision and parameter estimation, we can transform the parameters in the posterior probability of logistic regression model into solving a convex optimization problem with constraints.(3)By studying the MRF model for image segmentation, the improvement for the problem that MRF exists is the DRF model. The paper constructs a BLR-DRF model by studying the theory of DRF and BLR. This model is combined DRF model with BLR model. By obtaining the posterior probability distribution of the target image to determine the conditional probability of DRF, we build out a robust algorithm for image segmentation, considering the influence of the image noise and other factors.(4)In the experiment part, we combine the BLR model and the DRF model and applied to the field of the image segmentation of the pasture plant disease. According to the existing problem of disease images,we use the RGB and the HIS color space as the feature space and the R,G,B color values are selected to do the normalization process for the aim of reducing the impact of the light.Through the simulation experiment of our algorithm, we know that this hybrid model can deal with the noise and be better to keep the edge sharpness of the target,it has fast,reliable,effective and accuracy advantages, and work well for the plant disease spot and background segmentation.This model meets the needs of its further processing.