Wavelet multi-scale Markov network model-based image segmentation method

Abstract

The invention relates to the technical field of image processing, in particular to a wavelet multi-scale Markov network model-based image segmentation method. The method comprises the following steps of: intercepting training image blocks of an Nc type with uniform areas from an image to be segmented; training each type of training image blocks, constructing a characteristic vector of each position on the scale according to a position corresponding relationship of wavelet coefficients among frequency bands on the same scale, and estimating a Gaussian Markov model parameter on the corresponding scale by using a least square method; performing J layer wavelet conversion on the image to be segmented; calculating the multi-scale likelihood of the characteristic vector on each scale corresponding to each type of texture from the bottom to the top according to an initial scale parameter alpha of a given multi-scale characteristic network model, and estimating an interactive parameter alpha among scales from the top to the bottom; and establishing a multi-scale mark field network model, and segmenting the image. Through the method, a more accurate segmentation result can be acquired, andthe method can be used for segmenting a texture image and an aerial image.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an image segmentation method based on a wavelet domain multi-scale Markov network model, which can be used for the segmentation of texture images and aerial images. Background technique [0002] Image segmentation refers to the process of dividing an image into regions with different characteristics and extracting objects of interest. Image segmentation is a key step from image processing to image analysis, and it is also a basic computer vision technology. This is because image segmentation, object separation, feature extraction, and parameter measurement transform the original image into a more compact form, enabling higher levels of analysis and understanding. [0003] However, the purpose of image segmentation is for image understanding, but the ideal segmentation results often require the results of image understanding as prior knowledge, which brings great difficu...

AI Technical Summary

Problems solved by technology

[0008] The present invention overcomes the problem of insufficient local information statistics in the traditional method, provides an image segmentation method based on the multi-scale Markov network model in the wavelet domain, uses the advantages of wavelet transform to express the non-stationary characteristics of the image, and combines the positional relationship of wavelet sampling and Gaussian Based on the characteristics of the Markov random field (GMRF) model, a multi-scale feature network model is established; combined with the multi-scale random field model MSRF and the multi-scale Potts model, a multi-scale marker network model is established

Images