Performance Evaluation of a Deep Learning-based Optimal Image Compression Scheme in Two Distinct Wireless Models

Abstract

The need for dependable transmission of high-quality images across wireless communication networks has grown due to the quick expansion of multimedia applications, including web services, digital photography, and medical imaging. Wireless channels, however, are particularly vulnerable to multipath fading, which raises the Bit Error Rate (BER), causes data loss, and deteriorates system performance. Therefore, reliable transmission techniques and efficient image compression are essential to ensure high-quality image delivery in poor channel circumstances. In this regard, this manuscript proposes an effective deep learning-based image compression framework intended to maintain image quality during wireless transmission as a solution to this problem. This manuscript first examines an autoencoder-based optimum compression strategy. The proposed autoencoder compression scheme obtains PSNR values of 24.81 dB and 27.58 dB, low MSE values of 0.004 and 0.002, and compression ratios of 0.38 and 0.75, respectively, according to performance evaluation using the Microsoft COCO and CIFAR-10 datasets. Secondly, using different modulation techniques like QPSK and BPSK with and without diversity schemes, the compressed images are transmitted over simple and sophisticated wireless channel models, like the Rayleigh fading and Nakagami-m fading channel models, respectively. According to experimental data, diversity-assisted modulation achieves PSNR values between 43.39 dB and 55.18 dB, greatly improving visual fidelity and transmission dependability. The proposed method successfully strikes a balance between the effectiveness of compression and image quality, making it a viable option for wireless multimedia communication in difficult channel circumstances.