Removal Of Blocking Artifacts From JPEG-Compressed Images Using a Neural Network

The goal of this research was to develop a neural network that will improve the quality of JPEG compressed images, irrespective of compression level. After reviewing related articles, published papers, and previous works on developing a computationally efficient algorithm for reducing the blockiness and Gibbs oscillation artifacts in JPEG compressed images, we decided to integrate a neural network into a previously developed technique. For this approach, the Alphablend filter [35] was used to post process JPEG compressed images to reduce noise and artifacts. The Alphablend result was further improved upon by the application of a trained neural network. We compare our results with various other published works using post compression filtering methods.