Study On The Method Of Main Structure Extraction From The Texture Images Via Relative Total Variation

The main structure of the image is including the main information of the image,such as large-scale edge,meeting the characteristics of continuous smooth.The image texture part refers to some small-scale detail of the image,carrying background and detail information of certain characteristics,part of the texture is relatively smooth and have strong gradient.For many "main structure + texture" mixed images,In many cases,the extraction of large-scale edge parts is more valuable to the researchers.So how to extract the main structure part from the coexistence of the "main structure + texture"image is an important research direction.The technology not only can enhance the understanding of the image,but also can be used in remote sensing image multi-scale segmentation and edge detection.However,the "main structure + texture" image exists in a complex state,which is a challenge for extraction from the main structure.For this question,the current research methods mainly focused on the transcendental measure of the main structure and texture.But the requirements of the texture rules are also higher,For the random texture images can often not achieve satisfactory results.In this paper,we use the method of relative total variation model,By capturing the main structure and texture of the two types in the visual form of the essential differences,Remove the texture details of the image,achieving the purpose of the main structure extraction from the complex texture image.In this paper,the simulation of the sample image is carried out to verify the validity of the model in the extraction of the main structure.At the same time,in the process of validating the validity of the model,Add Gaussian noise to the original image to verify the model has a good performance in suppressing noise.In this paper,this model is applied to multi-scale segmentation and edge detection of remote sensing images and get a better result.