In this paper, we present the effectiveness of a frequency domain-based loss function using the discrete cosine transform (DCT) for the bit-depth enhancement (BDE) problem that recovers high-bit-depth images from low-bit-depth images. By minimizing the loss between the DCT coefficients, it is expected to recover smooth luminance changes by suppressing extra frequency components caused by the weak gradients contained in the false contour artifacts. Moreover, we proposed a frequency domain-based multi-level BDE method to deal with different bit-depth degradation. The proposed multi-level BDE method identifies the bit-depth of the input image by embedding the bit-depth information in the frequency domain, and it recovered the missing lower bits appropriately. Experimental results show that the model optimized with frequency-based loss outperforms the model optimized with other losses in the comparison considering the objective and subjective results. Furthermore, we show that the proposed multi-level BDE method is effective for more severe bit-depth degradation on a specific benchmark dataset.
The effectiveness of utilizing inter-channel correlation and self-similarity for demosaicking has been reported in many literatures. On the other hand, many convolutional neural network (CNN)-based demosaicking techniques have also been proposed to achieve state-ofthe-art accuracy. In CNN-based demosaicking, one of the most important issue is how to consider the correlations using neural network. In this paper, we propose a novel CNN-based demosaicking method that considers an effective combination of both inter-channel correlation and self-similarity. Specifically, we apply the CNN to predict the color differences R-G and B-G, then the demosaicked image is obtained from the predicted color differences and the input color filter array (CFA) image. At the same time, our network considers the self-similarity in the color difference domain by applying non-local attention for high-level feature map. Experimental results show that our method provides the better accuracy and visual performance compared with conventional demosaicking methods. In addition, the versatility of the proposed framework is demonstrated by experiments with images sampled by various CFA patterns.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.