Edge Detection of Welding Seam Image Based on Two-dimensional Fuzzy Entropy and Genetic Algorithm

Abstract

The weld image acquisition is generally under the environment of arc, soot and other complex, weld image edge detection is more difficult, this paper proposes a new two-dimensional fuzzy entropy extraction algorithm. When dealing with image edge detection, using the genetic algorithm to obtain the global optimal solution of the edge, the algorithm calculate the two-dimensional fuzzy entropy threshold, thereby which avoid the need large amount of calculation. The simulation results show that the genetic algorithm’ convergence to the optimal threshold, near won greatly reduce the running time, detection of weld edge image clearer, achieve the desired effect. Introduction As the visual sensor, the image processing algorithms and the development of intelligent control technology, using the optical visual sensing of weld tracking technology obtained the rapid development of weld edge is the basis of the seam tracking and quality control of welding process. In the process of welding, image sensors by arc and other factors, such as dust and noise interference, makes the conventional image processing result is not stable, so improve the weld edge detection precision has very important significance. Weld edge is the most important characteristics of weld image, realize the weld edge detection has been a hotspot of research on image processing, people expect to find a kind of strong noise, positioning, not leak, not mistakenly identified on the edge of the method. The method currently applied to weld image edge detection is operator edge detection method (Roberts operator, Sobel operator, log operator, Prewitt operator, Canny operator) ,morphological method, etc The fuzzy entropy method is used to measure the size of the image segmentation ambiguity, the image which is segmented by the fuzzy entropy contain the size of the original image information. In the actual image, due to the interference of noise or in the case of image contrast is not obvious, the distribution of the object and background with overlapping and indistinguishable, the distribution of the gray-level histogram image may not appear obvious bimodal or multimodal features, and therefore often get satisfactory segmentation effect, sometimes appear obvious segmentation error. Two-dimensional Fuzzy Entropy The previous method only use the image pixel gray level information, and not make full use of all the useful information in the image. A improvement of the problem which the space of the image information is introduced, the increase of image pixel point features, so as to construct a two-dimensional histogram. Objects and background in a two-dimensional histogram will be easier distinguished than in one-dimensional histogram. We know that each pixel in image and the correlation between neighborhood pixels is very big, make full use of the gray level information and spatial information of image segmentation will improve the effect of image edge detection. The two-dimensional gray histogram was constituted by each pixel gray value and its neighborhood grayscale average. Assuming a grayscale image I gray-scale series for L ,its size is * M N . ( 1, 2,.. , 1, 2,.. ) mn I m M n N = = said coordinates of pixel gray value ( , ) m n .Set T neighborhood average image is 3 * 3 in the image I , and The I and T can form a binary set International Conference on Intelligent Systems Research and Mechatronics Engineering (ISRME 2015) © 2015. The authors Published by Atlantis Press 703 * ( , ) {( , )} mn mn M N I T I T = , assuming that the bright area B Block is divided into the fuzzy region 1 B and fuzzy region 2 B , namely: