Research On Image Segmentation Method Combining Gradient And Non-local Means

Image segmentation is an image processing method that divides an image into several meaningful,non-overlapping regions according to certain similarity criteria.In the same area,the features of the image(such as grayscale,texture and color)are similar,while in different regions,the features of the image are significantly different.The threshold method has become the most commonly used image segmentation method because of its simple and practical advantages,but it only uses the gray-level information of the image,and in some cases,the ideal segmentation result cannot be obtained.Researches have showed that considering the gray-level information of image and the spatial correlation between pixels can improve the segmentation results,and introducing the spatial correlation between pixels into the threshold selection process can improve the segmentation results and improve the segmentation performance.In this respect,this paper uses the non-local means filtering and the local gradient mean ratio information of the pixels to construct a two-dimensional histogram,introducing the spatial neighborhood information between pixels into the threshold segmentation process through this method.Specifically,the following research work has been done:Firstly,a new gray-level-Ratio gradient two-dimensional histogram is proposed for the insufficiency of the traditional two-dimensional histogram which ignores the edge information of image.The gray-level-Ratio gradient two-dimensional histogram is composed of the local gradient mean ratio and gray-level of the pixels,which depicts the gray level variation of the pixels in the neighborhood.At the same time,the local gradient mean ratio is only related to the ratio itself and is independent of the mean value,so the local gradient mean ratio can be compared with a global threshold(T).The comparison of the two parameters together makes a new two-dimensional histogram---a two-dimensional histogram containing edge information about gray-level and mean Ratio---gray-level-Ratio gradient two-dimensional histogram.Based on this,this paper gives a two-dimensional minimum cross entropy threshold segmentation method,which achieved a good segmentation effect.In addition,a new non-local means two-dimensional histogram is proposed to solve the problem that the traditional gray-level-local means histogram will blur the image and lose more details.The method constructs a two-dimensional histogram together with the non-local means of the image pixel neighborhood and the gray level of the image.The non-local means filtering method breaks through the limitation of the traditional filtering method that only weights the average in the smaller local neighborhood of the image while maintaining more image structure information.Compared with local mean filtering,this method can reflect the discrete degree of image gray level and preserve the details of the image more completely.Then the advantages of relative entropy are analyzed.On this basis,the relative entropy threshold segmentation method based on gray-non-local mean histogram is proposed.The method achieves good results in the accuracy of image segmentation results.