Lossy Image Compression Approach Based Wavelets and Convolutional Auto-Encoder

doi:10.1007/978-3-032-02312-4_11

Return

Lossy Image Compression Approach Based Wavelets and Convolutional Auto-Encoder

2026-01-01

PRE

DOI：10.1007/978-3-032-02312-4_11

Original

Original request for help

Save

Abstract

En 中文

Up to now, numerous studies and research endeavors have explored the application of Discrete Wavelet Transform (DWT) in the field of image compression. Indeed, this mathematical tool, when combined with other algorithms, has proven its ability to reproduce images with high visual quality and less storage space. In this paper, we built a compression system that follow the same scheme as a baseline standard using DWT transform combined to a Convolutional Auto-Encoder (CAE) configuration to extract finer details from uncompressed images, a scalar quantization to quantify spectral coefficients and finally a Huffman code to get data binary code. The performance of this innovative model, referred to as DWT-CAE, was evaluated based on the Compression Ratio (CR), which is associated to the memory space occupied by the compressed image and the Structural SIMilarity (SSIM) factor. The experiments were conducted on Kodak dataset and obtained results indicate that the presented approach achieved an average SSIM of 0.83, producing images with clearer details and textures, while occupying an average storage space of 16,4%. To demonstrate the proposed approach strength, we conducted a comparative study that revealed that our novel system offers a gain of average space memory of 76% compared to the ordinary baseline algorithm and 7% compared to JPEG standard.

Keywords:

Discrete Wavelet Transform DWT

Image compression

Convolutional Auto-Encoder

Baseline algorithm

JPEG standard

Structural SIMilarity

Compression Ratio