LAUSR

Denoising convolutional neural networks (DnCNNs), initially developed for Gaussian noise removal, are powerful nonlinear mapping models in image processing. After changes in training data, they can be used for suppression of random-valued impulse noise (RVIN) with excellent results. To achieve favorable denoising performance, however, it is necessary to have an accurate perception of the noise ratio so that the most suitable DnCNN can be chosen for denoising. Thus, this model is severely limited in flexibility. To address this problem, we propose a blind CNN model for RVIN denoising with a flexible noise ratio predictor (NRP) as an indicator. Some patches are randomly selected from the RVIN-corrupted test image, and feature vectors that indicate whether the center pixel is contaminated or not are extracted by the predictor. These feature vectors are composed of multiple statistics, namely, the multiple rank-ordered absolute differences (ROADs), the clean pixel median deviation (CPMD), and the edge pixel difference (EPD). They are rapidly mapped to noise/clean (1 for noise, 0 for clean) labels by the pre-trained noise detector (the key component of our NRP). According to the ratio of the obtained noisy labels to the total number of selected patches, the predictor provides the noise ratio of the whole image. From the output of the NRP, i.e., the predicted noise ratio, the most appropriate DnCNN specifically trained for this noise ratio is exploited for denoising. Under the guidance of the NRP, the proposed method has the ability to handle unknown noise ratios. Simulation results indicate that our blind denoising CNN model achieves state-of-the-art performance in terms of both execution efficiency and restoration results.