Image Edge Detection Algorithm Based on The Parallel Genetic and Otsu dual-threshold

Abstract

The traditional genetic algorithm is used to search some function extreme value, but it has low precision and poor stability. For this shortage, A kind of parallel genetic algorithm based on Otsu double threshold algorithm of image edge detection, the design into a line of two columns of the vector population, than as an individual's fitness variance between, generated a number of initial population of parallel computing, At the same time, the largest fitness individuals in each generation directly copied into the next generation, evolution in a number of different subgroup in parallel, in order to avoid the premature convergence phenomenon appeared in the process of single population evolution, to improve the algorithm convergence speed. The experimental results show that the algorithm is compared with the traditional genetic algorithm, can more accurate and has better accuracy and stability. Introduction The edge is an important feature of image target detection, the image edge detection is a basic in the field of image processing and computer vision problems, and which is the premise of image subsequent processing. The edge detection can deal with the many complicated problems .How to find the real target image edge, and it has always been a hot research topic in the image processing field, the traditional edge detection methods have difference operator method (Roberts, Sobel, Prewitt, Kirsch and Laplacian, etc.)、template matching、optimal curved surface fitting, etc. [1].These algorithms have made some achievements in improving edge detection effect, However, they have complicated mathematical model and a longer run time also. Threshold value is a key method in the image edge detection algorithm, They are widely used because of simple, rapid and traditional Otsu method is a kind of the earliest by Otsu [2] in 1979, the classical threshold method is put forward, it is very effective and widely used [3]. This method uses one dimensional image gray histogram of the, make the exhaustive search pixels can be divided into two types of target and background, and the maximum variance between one dimensional threshold. Two-dimensional Otsu method [4-5] than traditional Otsu method has better noise performance. In practical applications, the threshold value method is usually as a basic segmentation technology which can obtain good image segmentation effect. But the threshold method only consider pixel gray value in the process of image segmentation, without considering the space characteristics, so sensitive to noise. The reason that the Otsu dual-threshold threshold is used, is the several threshold of the gray-level histogram can be divided into separate classes, making all kinds of the variance between the biggest. To determine the threshold value is the key of threshold segmentation. According to the number of threshold value, the image threshold can be divided into single threshold segmentation and the multiple threshold segmentation problem which can be converted into a series of single threshold segmentation problem to solve, but it will search a best threshold combination in the range of gray level, which encounter much more time consuming, difficult to practical application. To simplify the calculation, the available genetic and the immune evolutionary algorithm[6-7] which were used search the optimal threshold. Although the algorithm improves the speed of segmentation, the Otsu dual-threshold algorithm search result is not stable, thus get the threshold is not stable and accurate, lead to different image segmentation results. International Conference on Intelligent Systems Research and Mechatronics Engineering (ISRME 2015) © 2015. The authors Published by Atlantis Press 699 Because of these weaknesses, this paper designed image edge detection method combining with the OTSU threshold technology and the parallel genetic algorithm, on the one hand, using the parallel genetic algorithm to the high speed of parallel computer parallelism and parallel genetic algorithm, the combination of greatly improved genetic algorithm to solve the speed and quality, On the other hand, different subgroups are evolving in multiple parallel, the single population evolution phenomenon of premature convergence was avoided in the process, improve the rate of convergence of the algorithm. Simulation results show that the algorithm can accurately and quickly find the best threshold of image segmentation, and carries on the double threshold segmentation of image, through the best threshold edge detection and the result can be seen that the algorithm to improve the stability of algorithms in image edge detection and precision. The OTSU Threshold Method A grayscale range in the image for {0, 1,..., L 1}, the threshold value is set t ,which will be divided into two categories, the target image and the background image, the gray level’ probability is set i P , the probability of target parts is 0 0 t i i p π = = ∑ , the probability of the part background of is