Optimization for High-Efficiency Compound Image Compression Using Hybrid CNN-Huffman Encoding Framework with Minimum Cross-Entropy Method

Abstract Compound image compression is important in handling visual information consisting of a blend of graphics, text, and images, as seen in web pages and documents. This paper proposed a new Hybrid Convolutional Neural Networks-Huffman Encoding with Minimum Cross-Entropy Segmentation (CNN-HEMCE). CNN-HEMCE framework is a sophisticated image compression model that integrates deep learning, statistical segmentation, and entropy coding to effectively compress compound images without compromising quality. The method of choice starts with preprocessing procedures such as resizing, Gaussian smoothing, and normalization to improve the image clarity. The images are segmented by using the MCE method to segment major regions and then data augmentation procedures to enhance the generalization of the model. Feature extraction is done through a CNN that captures vital elements like textures, edges, and shapes. The last compression is obtained via Huffman encoding for minimizing data size without loss of critical information. The finding shows CR achieved 11.25, MSE attained 1.58, PSNR earned 48.36 dB, and SSIM attained 0.990. The proposed method effectively improves compression effectiveness and image quality, thus making it extremely suitable for complex and high-resolution compound image applications.