The Image Processing Technology Based On Energy Minimization Model

As a kind of the images technology, image processing is digitized by the computer so as tomeet the various needs of the image. It has been explored, innovated and improved by domesticand foreign scholars. Image segmentation is one of the most important image processingtechnology methods, and it gets the goal through various image algorithms from the image andmakes preparation for the further research work.The energy function model applied in the image segmentation is one of famous algorithms inthe mathematical models. It has become a widely used in image processing with high accuracy andefficiency. In this paper, the image processing technology about energy minimization models ismainly studied. Level set, C-V model, GraphCut and GrabCut algorithm which is based on graphcuts theory are reserched, as well as the application of this algorithm in the digital image andmatting. This article points of innovation as follows:First of all, a discrete Level set function algorithm based on graph cut is proposed. In view ofthe iterative process of the Level set and C-V model is easy to appear the problem of local minimain the energy function, combined with the advantage of iteration in GraphCut algorithm, dispersedthe energy function of C-V model which is accorded with the GraphCut algorithm, and gives theproof which is satisfied the diagram type theorem. The algorithm achieved a better segmentationeffect than C-V model in the segmentation of medical images, and improved the iterativeefficiency.Secondly, a matting algorithm of natural images based on HSV space and GrabCut isproposed. The GrabCut algorithm is an energy function model based on Gauss mixture model, andmainly used for digital matting. The result of new method depends on the initial parameter. Theimproved algorithm uses HSV spatial features of images to obtain the main color wave form,calculating parameters of the main color waveform to set the parameters of GMMS, and leadinginto an iterative differential energy function to optimize the iterations to improve the efficiency ofthe algorithm. The improved algorithm which applied to the natural images matting with fewcolors has better segmentation effect.Finally, a recognition algorithm based on improved Otsu with application inimmunohistochemical images is proposed. Immunohistochemical image is one of the differentialorgans such as the liver lesion and the maximum between class variance method (Otsu) is acommon method for the recognition to such images. However this method is prone to error in distinguishing negative areas. This paper proposed an optimization algorithm which combines thefeature information with hue and saturation components of the images can recognize the negativeregion better, experiments show that the algorithm has higher accuracy of segmentation than thesingle application ofthe Otsu.