QW-CNN: Secure and Robust Image Compression Using Quincunx Wavelet Decomposition, Lightweight CNN Modeling, and Selective Encryption over Noisy Channels

Abstract

With the rapid growth of digital imaging and multimedia applications, efficient image compression while ensuring data security has become a critical challenge. This paper proposes a novel framework, QW-CNN, which integrates Quincunx Wavelet Transform (QWT) for multiresolution image representation, a lightweight convolutional neural network (CNN) for adaptive coefficient prediction, and selective encryption to enhance security during transmission over noisy channels. The proposed method compresses images by predicting and thresholding wavelet coefficients using the CNN, followed by encrypting only high-magnitude coefficients to reduce computational overhead while maintaining confidentiality. Huffman-based entropy coding is applied to further reduce data size. The robustness of the framework is evaluated under additive white Gaussian noise (AWGN) channels. Experimental results on standard benchmark images (Lena, Pepper, Mandrill, Pirate, and Cameraman) demonstrate that QW-CNN achieves high compression efficiency, excellent visual quality, and strong security performance. The proposed method consistently delivers peak signal-to-noise ratio (PSNR) above 40 dB and structural similarity index (SSIM) above 0.97 for SNR levels above 40 dB, highlighting its effectiveness in secure and reliable image transmission..