Journal of Electrical Engineering and Renewable Energies Development
Volume 2, Numéro 3, Pages 107-115
2026-10-12

Efficient Medical Image Compression Using A Cnn And Discrete Wavelet Transform



Authors : Chikhaoui Dalila . Beladgham Mohammed . Ben Aissa Mohamed .

Abstract

This study presents a computationally efficient framework for medical image compression that integrates a low-complexity convolutional neural network (CNN) with the discrete wavelet transform (DWT). Unlike conventional CNN–DWT hybrid approaches that embed wavelet processing within large end-to-end architectures, the proposed method maps the input image to a compact three channel latent representation that is encoded using the CDF 9/7 wavelet, as employed in JPEG 2000. This design avoids custom coding of high-dimensional feature tensors while facilitating compatibility with established image-coding workflows. The CNN reduces spatial redundancy while preserving essential image information, whereas the DWT exploits frequency-domain sparsity and multiresolution characteristics to improve compression efficiency. A two-stage training strategy is adopted: the encoder–decoder is first pretrained without the codec and subsequently fine-tuned with real non differentiable quantization in the compression loop, enabling optimization against the actual coding process. Experiments on a publicly available low-grade glioma (LGG), magnetic resonance imaging (MRI) dataset demonstrate compression ratios of up to 40:1, with a peak signal-to-noise ratio (PSNR) of 38.6 dB and a multiscale structural similarity index measure (MS-SSIM) of 98.84%. The proposed CNN contains approximately 0.26 million trainable parameters and can be trained on a single GPU, demonstrating its potential for resource-constrained medical image storage and transmission applications.

Keywords

Image compression, convolutional neural network (CNN), discrete wavelet transform (DWT), medical image

Les articles similaires

Forecasting Of Co2 Emissions In Algeria Using Discrete Wavelet Transform –based Autoregressive Integrated Moving Average Models

Sahed Abdelkader .  Mékidiche Mohammed .  Kahoui Hacene . 
pages 443-449.

The Relationship Between Futures Contracts And Futures Indicators According To The Discrete Wavelet Transform Method For Maximum Overlap In The Saudi Financial Market

Alia Abdelbasset Abdessamad . 
pages 1307-1324.

The Fast Fourier Transform And The Continuous Wavelet Transform Analysis Of The Normal And Pathologicals Phonocardiogram Signals

Debbal S.m .  Bereksi-reguig F . 
pages 81-86.

Lossy Image Compression Using A Three Step Nonlinear Wavelet

Terki Nadjiba .  Bentrah Wafa .  Saigaa Djamel . 
pages 27-32.

Modeling And Classification Of Lithofacies Using The Continuous Wavelet Transform And Neural Network: A Case Study From The Berkine Basin (algeria)

Ouadfeul Sid Ali .  Zaourar Naima .  Boudella Amar .  Hamoudi Mohamed . 
pages 103-118.

Formats de citation