Two-dimensional Minimum Error Segmentation Method Based On Mean Absolute Deviation From The Median

Image segmentation is the basic research content in the area of image engineering, and it has the extremely important influence for analysis, understanding of image. Threshold segmentation algorithm based on gray level histogram of image information to selecting the optimal threshold by the threshold criterion. It is a simple, fast and effective segmentation method, and has widely used in the field of image segmentation, and the segmented image has a better quality than the segmented image which use the Otsu method and the maximum entropy method.The minimum error method is a classical threshold segmentation method which is proposed based on the template matching, in an image, the proportion of target and background size has little influence on the segmentation results. But when the actual digital image histogram has a very difference to the Gauss mixture distribution model, the results of segmentation threshold will be shift and segmentation accuracy will be reduced; For the traditional two-dimensional minimum error method, it has a poor ability to suppress noise, so it has a poor image segmentation effect to the images which has a low signal noise ratio. Combined with the existing threshold segmentation results, this article do some research and improvement of the minimum error method, the main contents in this paper include those aspects:(1) The minimum error method based on mean absolute deviation from the median is proposed. For the neighbor gray level information of pixel in the image, the median information in mean absolute deviation from the median is a more robust estimator of gray level than the mean information in variance, using the mean absolute deviation from the median as a discrete measure can get a better performance in suppress image noise; Based on the prior knowledge of the noise and background area in two-dimensional histogram model, in order to improve the computational speed, this two dimensional algorithm has been decomposed into two one-dimensional algorithms, and has achieve more accurate segmentation results and more robust performance on image which has the skew and heavy-tailed distribution.(2) The conventional two dimensional histogram usually use a direct divisionmethod to divide the area of noise, target and background, in the process of image segmentation is easy to lose the edges between the target and background information. Slash division method can overcome this shortcoming to get a better segmentation result. A two dimensional histogram structured by the neighbor gray level information of mean and median is proposed, and this segmentation algorithm has a better inhibition effect to the mixed noise which contents the Gauss noise and the salt and pepper noise.(3) In practical applications, due to the uneven lighting image is difficult to get a good segmentation result by using global threshold segmentation algorithm. Combined with mathematical morphology knowledge, a new segmentation algorithm based on the minimum error method is proposed. By the early treatment of mathematical morphology to equalize the image illumination, the algorithm have certain effect on the uneven lighting image segmentation.