Research On Image Segmentation Optimization Algorithm Based On Chan-vese Model

Image segmentation is one of the most challenging tasks in image analysis and pattern recognition.The importance of segmentation result has a direct impact the effectiveness of the continuous task,and it plays a vital role in image processing.Although the image segmentation technology has been rapid development,but there is no one way to all types of image segmentation In this thesis,the segmentation is mainly based on the CV model,using intelligent heuristic optimization algorithm for geometric image segmentation and other optimization algorithms for artifacts and illumination bias image segmentation.The main innovations of this thesis are as follows:(1)This thesis proposes an efficient bi-convex fuzzy variational method with teaching learning based optimization(TLBO)for geometric image segmentation.Firstly,this thesis adopt a bi-convex object function to process geometric image.Then,adaptively selecting the value of the length penalty item by the two stages of studying from teacher and learning from each other.The optimal region is searched firstly,then finding out the single optimal solution from the range.At last,efficiently combining the length penalty item and the numerical remedy method to get better results.Experimental results show that this thesis's method is effective than the existing methods.(2)Region-based image segmentation has essentially been solved by the Chan-Vese(CV)model.However,this model fails when images are affected by artifacts and illumination bias that outweigh the actual image contrast.Here,this thesis introduce a model(CV+XB)for segmenting such images.Firstly,in the energy function of the CV model,a binary artifact label X,is introduced to minimize the energy function through the dynamic threshold.This thesis proposed a dynamic artifact class preventing intensity artifacts from skewing the segmentation.Secondly,in retinex fashion,the thesis decompose the image into a piece-wise constant and illumination bias.The segmentation is parameterized using a phase-field,and efficiently minimized using threshold dynamics.The thesis demonstrates the proposed model on a series of sample images from diverse modalities exhibiting artifacts and/or bias,and the convergence of our algorithm is particularly high.