Lossy Image Compression Using Hybrid Deep Learning Autoencoder Based On k-mean Clustering

Abstract

Image compression plays an important role in reducing the size and storage of data while increasing the speed of its transmission through the Internet significantly. Image compression is an important research topic for several decades and recently, with the great successes achieved by deep learning in many areas of image processing, especially image compression, and its use is increasing Gradually in the field of image compression. The deep learning neural network has also achieved great success in the field of processing and compressing various images of different sizes. In this paper, we present a structure for image compression based on the use of a Convolutional AutoEncoder (CAE) for deep learning, inspired by the diversity of human eyes' observation of the different colors and features of images. We propose a multi-layer hybrid system for deep learning using the unsupervised CAE architecture and using the color clustering of the K-mean algorithm to compress images and determine their size and color intensity. The system is implemented using Kodak and Challenge on Learned Image Compression (CLIC) dataset for deep learning. Experimental results show that our proposed method is superior to the traditional compression methods of the autoencoder, and the proposed work has better performance in terms of performance speed and quality measures  Peak Signal To Noise Ratio (PSNR) and Structural  Similarity Index (SSIM) where the results achieved better performance and high efficiency With high compression bit rates and low Mean Squared Error (MSE) rate the results recorded the highest compression ratios that ranged between (0.7117 to 0.8707) for the Kodak dataset and (0.7191 to 0.9930) for CLIC dataset. The system achieved high accuracy and quality in comparison to the error coefficient, which was recorded (0.0126 to reach 0.0003) below, and this system is considered the most quality and accurate compared to the methods of deep learning compared to the deep learning methods of the autoencoder