Deep image denoisers achieve state-of-the-art results but with a hidden cost. As witnessed in recent literature, these deep networks are capable of overfitting their training distributions, causing inaccurate hallucinations to be added to the output and generalizing poorly to varying data. For better control and interpretability over a deep denoiser, we propose a novel framework exploiting a denoising network. We call it controllable confidence-based image denoising (CCID). In this framework, we exploit the outputs of a deep denoising network alongside an image convolved with a reliable filter. Such a filter can be a simple convolution kernel which does not risk adding hallucinated information. We propose to fuse the two components with a frequency-domain approach that takes into account the reliability of the deep network outputs. With our framework, the user can control the fusion of the two components in the frequency domain. We also provide a user-friendly map estimating spatially the confidence in the output that potentially contains network hallucination. Results show that our CCID not only provides more interpretability and control, but can even outperform both the quantitative performance of the deep denoiser and that of the reliable filter, especially when the test data diverge from the training data.
In this paper, a convolutional neural network for joint image demosaicing, denoising, deblurring, super-resolution and clarity enhancement is proposed. The network inputs are four-channel Bayer CFA image (R, G, G, B) and three channels of the same size containing distortions maps, namely, noise level map, blur level map, and clarity degradation map. It is shown that the designed network FiveNet can effectively process images with the mix of five different distortions. It is also demonstrated that adding clarity enhancement into the processing chain can additionally increase image quality (by up to 3-4 dB in PSNR). A small dataset ClarityDegr120 of color images with different clarity degradations and enhancements is designed using images processed by FiveNet. Mean opinion scores (MOS) for the test set are collected. The MOS prove that clarity enhancement can significantly increase image visual quality. A comparative analysis using the MOS demonstrates a low correspondence between image quality metrics and human perception for the clarity enhancement task.