A HYBRID DEEP LEARNING SYSTEM WITH DISCRETE HERMIT WAVELET TRANSFORM FOR IMAGE COMPRESSION

Abstract

This work presents a hybrid system combining convolutional neural 

networks (CNNs) with discrete Hermit wavelet transform (DHWT) to compress 

color images with high accuracy and efficiency. This is achieved through 

mathematical derivation of new waves from Hermit polynomials. The proposed 

waveforms possess properties such as orthogonality and convergence, enabling 

them to act as the primary factor in the multi-parametric analysis during initial 

processing. The CNN plays a crucial role in decoding, optimizing parameter 

quantization, and entropy encoding. The orthogonality of the proposed 

waveforms facilitates image compression, thus enabling the image to be 

processed by the proposed hybrid system.

The results obtained with the DHWT-CNN hybrid system demonstrate 

superior efficiency compared to conventional wave-based methods, with error 

ratios [missing value] and a peak signal-to-noise ratio (PSNR) of 48.72, 

resulting in a compression ratio of 52.3% while maintaining image quality. This 

confirms the efficiency of the proposed framework