Bilateral Filtering-based Texture Removal Algorithm With Keeping Special Details

The main content of this paper is the removal of textures in texture images.Texture removal is a common method of image preprocessing.For texture images,texture information that is not important to the semantics can be regarded as noise.Therefore,texture removal is also named as texture filtering.The core task of texture filtering is to remove the image texture and preserve the image structure information.However,textures and structures in texture image are mixed together.Therefore,removing textures without blurring the structure is a difficult problem,and it is also a research hotspot.However,existing image texture filtering algorithms have some disadvantages in removing image texture while keeping some special details.Aiming at the shortcomings of existing texture algorithms,our paper proposes a new texture removal algorithm.The algorithm aims to remove the texture information in the texture image while keeping the image structure information as much as possible,especially some special details,such as thin and long structures and corners,which are easily smoothed.The main contributions of this paper are as follow.First,a multi-directional relative total variation model for distinguishing image textures and structures are proposed;and combining the long structure and the relative total variation model based on multiple directions to estimate the texture filtering kernel scale.Based on the existing research,this paper proposes a new texture removal algorithm.This algorithm can well keeping the structural information of the image when removing the image texture,especially the special details such as the thin and long structure and corner information.First of all,this paper will use a structure detection method that can identify thin and long structures,use directional filtering on the texture image,standardize the gradient,and use the message transfer method to accumulate the normalized gradient along the gradient direction,and enhance the gradient in the continuous direction.Identify long structures and use them in subsequent texture filtering.Secondly,the original relative total variation model is improved,and the relative total variation model based on multiple directions is proposed.For the corner structures that are easily missed by the relative total variation model,the method of calculating the relative total variation in multiple directions is used,and the direct use of the relative total variation cannot accurately estimate the kernel scale,and the right and left neighborhood strategy is used to distinguish the distance from the boundary to accurately estimate the filtering kernel size,and the filtered kernel scale is estimated to generate a guided filtered image.Finally.the texture-removed image is obtained by combining the joint bilateral filters.Experiments show that the texture filter designed in this paper performs better in keeping special details such as thin and long structure,and sharper at the corner details of the image after the texture removal.Also,iterative convergence speed is faster.Finally,this paper also introduces the practical applications of texture filters in image detail enhancement,edge detection,image abstraction,and image segmentation.