Moiré Pattern Removal From Screen-shot Images Via Image Decomposition

In the process of the digital camera captured the image,the aliasing of the frequency will occur and appear as moiré in vision,when the sampling rate of the image sensor is close to the scene information frequency.The appearance of moiré heavily disturbs the qualitative and quantitative analysis of images.Since moiré patterns are extremely sensitive to the slight transformation of repetitive structures,diverse image structures mean the complexity and irregularity of moiré.Therefore the demoiréing research is very challenging.The traditional filtering methods such as Gaussian filtering and bilateral filtering are not effective in suppressing moiré patterns and there is little research on demoiréing.In order to solve this problem,this paper proposes a novel screen-shot images demoiréing method by guided filtering on wavelet domain and polyphase layer decomposition,which removes moiré patterns from screen-shot images effectively.The achievements and innovations of the paper are as follows:1.Based on the characteristic difference between the image information and moiré patterns,this paper proposes a band interception method in wavelet domain.Utilizing threshold interception of wavelet coefficients in the first level high-pass subband of wavelet domain,this method remove moiré patterns while maintaining the image details and ensuring the image sharpness.2.Based on the structural similarity among color channels and the feature of minimal aliasing in green channel,this paper proposes a guided filtering method in wavelet domain.Using the green channel image,this method restores the red and blue channel images,which avoids image smoothing applying guided filtering in spatial domain.3.Based on the structural differences between screen-shot moiré patterns and the image background,this paper proposes the polyphase layer decomposition based on Gaussian Mixture Model(GMM)method.The layer decomposition model separates the image into a background layer and a moiré layer,which are both regularized by a patch-based GMM prior;Due to the similarity of moiré patterns and the image details,this method employs a polyphase decomposition model to disturb the feature of moiré patterns while nearly preserving the feature of natural images.It avoids blocking effect and reduces the time complexity.The experimental results show that the proposed method demoiré efficiently on non-ground truth images and synthetic ground truth images.Comparing with stateof-the-art methods,the method have the best performance in the field of screen-shot images demoiréing.