Deep Architectures for Image Compression: A Critical Review

Deep learning-based networks have achieved great success in the field of image super-resolution. However, many networks do not fully combine high-level and low-level information, and fuse local and global information. A multiscale recursive feedback network (MSRFN) for image super-resolution is proposed. First, multiscale convolution is integrated into the feedback network to propose multiscale projection units that adaptively capture image features of different scales by driving a multipath information flow. Next, recursive learning is applied to multiscale projection groups composed of up- and down-multiscale projection units to construct a feedback module that exploits high-level information to correct the low-level representation and refines the features in the early layers. Then, global residual learning and local residual feedback were combined to provide more contextual information for the final reconstruction. Experimental results demonstrate that MSRFN can predict more high-frequency details and alleviate the ringing effect and checkerboard artifacts inherently in CNN-based models. Even when the training datasets are relatively small, MSRFN is still superior to most state-of-the-art methods, especially for large scaling factors ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\times 8$ </tex-math></inline-formula> ). 

/pdf/multiscale-recursive-feedback-network-for-image-super-30y2as9w.pdf

Multiscale Recursive Feedback Network for Image Super-Resolution

Deep learning-based networks have achieved great success in the field of image super-resolution. However, many networks do not fully combine high-level and low-level information, and fuse local and global information. A multiscale recursive feedback network (MSRFN) for image super-resolution is proposed. First, multiscale convolution is integrated into the feedback network to propose multiscale projection units that adaptively capture image features of different scales by driving a multipath information flow. Next, recursive learning is applied to multiscale projection groups composed of up- and down-multiscale projection units to construct a feedback module that exploits high-level information to correct the low-level representation and refines the features in the early layers. Then, global residual learning and local residual feedback were combined to provide more contextual information for the final reconstruction. Experimental results demonstrate that MSRFN can predict more high-frequency details and alleviate the ringing effect and checkerboard artifacts inherently in CNN-based models. Even when the training datasets are relatively small, MSRFN is still superior to most state-of-the-art methods, especially for large scaling factors ( $\times 8$ ).

https://ieeexplore.ieee.org/ielx7/6287639/9668973/09678324.pdf

In the era of intelligent media, we all need high quality image or video information. However, due to the limitation of bandwidth or storage resource, we usually use low bit rate coding technology for image compression. But the image distortion caused will bring in a visually inferior experience, and the actual information obtained is often not sufficient. Therefore, the super-resolution reconstructed image coding technology is researched to provide a higher dynamic bit rate variation range. Based on image super-resolution reconstruction, this paper proposes two targeted image compression coding frameworks. Firstly, we compress the DCT down-sampled image as the input of super-resolution reconstruction, and use image completion as the pretreatment process of image super-resolution reconstruction, which effectively improves the quality of reconstructed image at low bit rate and, hence improves the coding efficiency. Secondly, a new image compression technique is proposed here which is based on residual compensation and super-resolution reconstruction. In the encoder, the super-resolution reconstructed image is subtracted from the original image to obtain the residual image which is compressed. In the decoder, the residual image is decompressed and added to the reconstructed image to complete the decoding process which achieves flexible rate control.

Wei Zhong

Papers

Image Compression Algorithms Based on Super-Resolution Reconstruction Technology