Application Of Probabilistic Pseudo Morphology In Image Processing

Probabilistic pseudo morphology(PPM)is proposed when mathematical morphology(MM)is difficult to generalize to color images.Due to the lack of uniform standards for multivariate data ordering,and current multivariate data ordering are difficult to meet human visual needs,image processing results of mast of color morphologies based on sorting are not ideal.PPM considers from the perspective of statistical characteristics,using the pseudo extreme calculated by Chebyshev inequality to define basic operations,it don't need to consider multivariate data sorting problems,not only improves the anti-noise ability of the algorithm,but also reduces the amount of computation.This paper expounds the basic principle of PPM in grayscale images.It mainly discusses the condition of the value of parameter k and shows the parameter k makes the method have both linear and nonlinear characteristics in the image processing.Focusing on the problem of parameter k when probabilistic pseudo extremes are approaching the actual extremes,this paper proposes estimating the parameter k on the regularized histogram,making the error minimal on the basic operational results of PPM under the same structural element and the results of MM.While PPM extended to color images,we focus on the application of principal component analysis in global and local data.Comparing the estimation of fractal dimension of texture images with PPM and MM,and the result of dilation and erosion of the two methods,our method can better describe the texture features and effectively maintain the structure and preserve the texture of the images.Against probabilistic pseudo morphology only selecting the direction of the first principal component for operation,“distortion” occurs while the parameter is large.We propose an improved method in the framework of PPM by redefining the reference axis.Using the improved method constructing the open-close-close-open filter greatly increases the ability of the original probabilistic pseudo morphology in removing Gaussian uncorrelated noise and reduce "distortion".Comparing with other morphological methods combined with filtered error indicators.The framework of probabilistic pseudo morphology can effectively remove both types of noise.This paper also applies the PPM to image feature extraction,mainly for image edge detection and texture segmentation.Using probabilistic pseudo morphological gradient operation extract the edges of color images.Compared with MM,the edge results of our method are enhanced and the edge contours are more distinct.In the texture segmentation,the normalized morphological covariance is used to describe the texture features to obtain eigenvectors of each point.And then K-means clustering is used to complete the texture image classification.Our method can accurately classify foreground objects and has good ability to capture complex textures and scale changes.