Image Compression with Recurrent Neural Network and Generalized Divisive Normalization
This work addresses image compression for applications requiring high-quality reconstruction, representing an incremental improvement with novel blocks and techniques.
The paper tackles image compression by developing novel analysis and synthesis blocks using convolution layers and Generalized Divisive Normalization, along with a pixel RNN for quantization and LSTM for residual encoding, resulting in a variable-rate framework that outperforms existing methods and standard codecs like JPEG in image similarity.
Image compression is a method to remove spatial redundancy between adjacent pixels and reconstruct a high-quality image. In the past few years, deep learning has gained huge attention from the research community and produced promising image reconstruction results. Therefore, recent methods focused on developing deeper and more complex networks, which significantly increased network complexity. In this paper, two effective novel blocks are developed: analysis and synthesis block that employs the convolution layer and Generalized Divisive Normalization (GDN) in the variable-rate encoder and decoder side. Our network utilizes a pixel RNN approach for quantization. Furthermore, to improve the whole network, we encode a residual image using LSTM cells to reduce unnecessary information. Experimental results demonstrated that the proposed variable-rate framework with novel blocks outperforms existing methods and standard image codecs, such as George's ~\cite{002} and JPEG in terms of image similarity. The project page along with code and models are available at https://khawar512.github.io/cvpr/