An Artificial Neural Network Based Pixel-by-Pixel Lossless Image Compression Method
Sinem Gümüş, Fatih Kamışlı · 2022 30th Signal Processing and Communications Applications Conference (SIU) · 2022
With the success of deep learning in computer vision applications, deep learning based algorithms have also been proposed for image processing applications. One such application is lossless image compression. Most traditional lossless image compression algorithms use pixel-by-pixel processing algorithms. In the pixel-by-pixel algorithms, each pixel is predicted from the previously coded neighbor pixels and the prediction error is compressed without loss. Deep learning based lossless image compression algorithms can be categorized into two categories, namely prior based algorithms and pixel-by-pixel (or masked convolution based) algorithms. In the pixel-by-pixel algorithms, each pixel’s probability distribution is obtained by processing the previously coded neighboring pixels with a neural network, which is then used by an arithmetic coder for lossless compression. This paper explores a deep learning based architecture, which utilizes masked convolutions, to model probability distributions of pixels and also presents a method to improve the parallelization of the algorithms. The obtained compression performance is competitive and is compared to both state-of-the art traditional and deep learning based methods.