Research Of Color Image Edge Detection Based On Particle Filter

Edge detection is the based operation of most computer vision problem and image processing problem, not only simplify the image analysis by significantly reducing the amount of information, but also retain the structure of the useful information of scene. Now there are a lot of research on gray image edge detection,and the study of the color image edge detection is less, but the study found that gray image edge detection method less than color images edge detection method detecting the edge of about the one over ten. This paper proposes a new color image edge detection method, research of color image edge detection based on particle filter.In order to improve the algorithm of noise resistance, first get multiple scales image based on Gaussian smoothing. This paper makes the objective image respectively with three kernel convolution get three different scales of Gaussian smoothing image. Through the analysis of color image color space, this article choose RGB space decomposition and the use of synthetic method to detect image edge.Edge detection can be understood as the process of the edge points tracking process, particle filter algorithm is very effective to the object tracking problem. According to the theory of particle filter algorithm, this paper divides the image into K * K blocks, and choose starting points in the block. The establishment of the model is the important and hard part of the particle filter algorithm, design good state transition model and observation model. According to the characteristics of the edge, using normal distribution function and the curvature functions to collect sample points namely, to establish state transition model. Choose local contrast features, edge intensity, edge smoothness features as the observation target. Based on the histogram of image local contrast characteristic statistical analysis to determine the corresponding likelihood function for index form; According to the statistical analysis to determine the brightness on the edge of the image standard features and the smoothness of likelihood function of normal distribution form, adjust their expectation and variance values change their edge in determining the weights of the influence of size, final observation function is determined. Edge of the end of thetrack conditions identified as: track to known edge, sample point value is less than a certain threshold TL.Adjust various parameters in experiment, respectively for the three color image and their noise of the image edge detection, the test results with the classic operator sobel operator, LOG operator, Canny operator compares and analyses the test results in this paper, the test results of the algorithm is better.