Flat and Texture Region Segmentation in Image Denoising
Sung-Min Woo · 2022
Existing image denoising methods have shown to be effective in suppressing noise, however, they over-smooth edge and texture details excessively because it is difficult to distinguish signal and noise while denoising. In this paper, we propose a method to segment an image into texture and flat region for enhancing the performance of denoising. To do this, texture map is generated using Sobel gradients, and is utilized in the loss function of the proposed network to learn the residual of the flat regions. The proposed method can control the level of noise in the texture region while suppressing the noise in the flat region as much as possible.