Research On Image Segmentation Algorithm 30

Image segmentation is a process of dividing an image into different regions such that each region is, but the union of any two adjacent regions is not homogeneous. Image segmentation is the first step in image understanding and pattern recognition. The basic development process of the image segmentation is as follows: the early classical segmentation methods based on image intensity and gradient, such as threshold, contour-based, and region-based; the eighties active contour models, such as parametric active contours and geometric active contours; the segmentation methods with prior knowledge, such as active shape model and active appearance model. From the development process, we can observe that the intelligence and ability of the image segmentation become better and better.We analyse image segmentation as state estimation in the condition of non-Gaussian and on- linear. Particle filer is one of the powerful tools for non-Gaussian and nonlinear problem, and has been widely used in many time series problems, such as target tracking, signal processing etc. In recent years, some scholars have extended the application area. Patrick solved edge tracking based on Particle filter. Infrared target extraction and target contour extraction are traditional problems in image processing. The thesis focuses on how to solve based on the frame of Particle Filter.The infrared target extraction algorithm based on Particle Filter aims to the target extraction problem in infrared images of low SNR. We analyse and solve the problem of infrared target extraction in the view of state estimation, and compute the threshold value adaptively by approximately optimal estimation of a dynamic system. In the framework of Particle Filter, the threshold state space is established on the Gray-Variance Weighted Information Entropy and the gray value of each pixel. Particle Swarm Optimization is introduced to construct the state transition model. As for the observation model, a novel objective function, integrating gray, entropy, gradient and spatial distribution of pixels, is proposed for both the quantitive evaluation of the target extraction and the weight of each particle in the particle set. Finally, the estimation for target extraction threshold is the weighted average of all the particles. The Experimental results the effectiveness of the proposed algorithm.The contour extraction algorithm based on Particle Filter is designed for the contour extraction in color images. The contour extraction problem can be formulated as non-Gaussian/on-linear state estimation and solved in Particle Filter framework. In this thesis, the contour which is usually continues curve, is approximated by a set of lines of element length. Particle Filter is used for the estimation of lines'parameter. In this thesis, we construct the prediction contour based on edge detection operator or color space cluster algorithm, in order to guide getting the solution of contour extraction problem. Implementation of Particle Filter requires the state space model, the likelihood function of proposal distribution, the state transaction model, the measurement model and the resample method. In the contour extraction algorithm, we propose state space on the line's slope and intercept. The PSO state transaction model is used for particles'evolution. The measurement model is the key in Particle Filter framework. In this algorithm, we used the energy function, which is used for measure the validity of the contour in the Snake model, to calculate each particle's weight. The estimation of the state is given by weighted mean of the particle set.In this thesis, we focus on four problems as follows:(1) The state space model is the base of Particle Filter framework, which has key influence to the efficiency of Monte Carlo sampling. In the infrared target extraction problem, the threshold state space is established on the Gray-Variance Weighted Information Entropy and the gray value of each pixel. In the contour extraction problem, we propose state space on the line's slope and intercept.(2) The construction of proposal distribution. In this thesis, we construct the prediction contour based on edge detection operator or color space cluster algorithm, in order to guide getting the solution of contour extraction problem.(3) The construction of state transaction model. We adapt PSO state transaction function in both infrared target extraction and contour extraction algorithm, in order to improve the distribution of particles and promote the iterative convergence.(4) The construction of measurement model. a novel objective function, integrating gray, entropy, gradient and spatial distribution of pixels, is proposed for both the quantitive evaluation of the target extraction and the weight of each particle in the particle set. In the contour extraction algorithm, we used the energy function, which is used for measure the validity of the contour in the Snake model, to calculate each particle's weight.