Rail Image Segmentation Based On Neighborhood-Mean Weighed Valley-Emphasis Enhancement Algorithm

Since rail images are normally uneven gray-scale distributed, both Otsu algorithm and VEM algorithm are therefore difficult to recognize the defects from the background. This paper proposed a VEM (valley-emphasis enhancement) algorithm based on neighborhood mean weighting. First of all, according to the image histogram, calculate the effective traversal interval, and then divide the image into two types by using threshold histogram, and the between-class variance. Secondly, calculates the threshold of neighborhood average probability, with the neighborhood average weighted variance between classes. Generally, in the most efficient traverse range obtained between-class variance and the corresponding threshold value is the optimal threshold, that not only ensure maximizing variance between images, but also meet the best threshold value fell in the bottom of the histogram. Experimental results demonstrate that the algorithm proposed in this paper can segment rail images effectively, which is superior to Otsu algorithm and VEM algorithm in image segmentation performance, and has a certain degree of improvement on anti-noise. It has better image segmentation performance for histogram images with single peak distribution and double peak distribution, and expands the application scope of the algorithm.