Learning Recursive Filters for Low-Level Vision via a Hybrid Neural Network

View Researcher's Other Codes

Disclaimer: The provided code links for this paper are external links. Science Nest has no responsibility for the accuracy, legality or content of these links. Also, by downloading this code(s), you agree to comply with the terms of use as set out by the author(s) of the code(s).

Authors Sifei Liu, Jin-shan Pan, Ming-Hsuan Yang
Journal/Conference Name ECCV
Paper Category
Paper Abstract We specify more general settings and performance of the proposed network in Section 2. We visualize and analyze more weight maps generated by the deep CNN in the proposed algorithm in Section 3. We demonstrate the effectiveness of the proposed image denoising model via more quantitative and qualitative results in Section 4. In Section 5, we compare the filtered results by the proposed algorithm and the original implementations for edge-preserving smoothing and image enhancement. In Sections 6 and 7, more examples of image pixel interpolation are presented with comparisons to several state-of-the-art algorithms. We further provide an interesting application for color interpolation in Section 8. For ease of comparisons, we show all results by different methods in one page, and the details can be clearly viewed at the original image resolution, or equivalently, by zooming in on each figure. Meanwhile, we also create a video demo that combines the proposed RTV filtering through approximation, as well as the generated edge maps (See Fig. 2) to formulate the cartooning effect. The lightening of all frames are uniformly adjusted before LRNN processing, and the edges are directly summed to the filtered frames. The video is processed frame-by-frame, but is quit stable over the temporal domain in both filtering effect and edge producing.
Date of publication 2016
Code Programming Language MATLAB
Comment

Copyright Researcher 2021