High Quality JPEG Compression with CNN Based Pre-Editing and DE- Blocking

Abstract

Picture compression is one of the most basic requirement in any image processing systems. Unfortunately, anytime a loss approach is used, a large number of artefacts are produced. Is it always possible for us as a group to reach the requisite bit rate and quality? No, it is not correct. Because if we try to lower the bit rate, the image quality worsens. Over the years, many projects have been launched to fight this problem, with varied degrees of success. We presented a CNN-based pre-editing network architecture for improving compressed image quality while lowering bit rate. We hope to break the unholy relationship between bit rate and image quality in this study, as well as give a method for avoiding compression artefacts. The editing is done with a convolutional neural network. The loss between the input and the edited and compressed output is also evaluated in order to convert it to a lossless form. To demonstrate the technique, JPEG compression is utilized, which results in a reduction in bits and an improvement in quality.