Coiflets, artificial neural networks and predictive coding based hybrid image compression methodology

Abstract

Hybrid image compression system is discussed and analyzed for better objective fidelity metrics combining the advantages of Coiflet filter functions of wavelets, Predictive Coding (Differential Pulse Code Modulation-DPCM) and neural networks in addition to quantization and Huffman encoding techniques to eliminate the interpixel, psychovisual redundancy and coding redundancy. Artificial neural networks are self adaptive i.e. they can adjust themselves to data without any specification of the functional model they are fault tolerant by architecture. Wavelets (by choice) on the other hand are computationally simple and provide good compression ratios for high resolution images especially while DPCM removes redundancy in the information. Initially selected wavelet of choice (Coiflet5 in this case) is applied on the input image for two level decomposition generating seven bands of low frequency and high frequency coefficients. The low frequency band 1 coefficients are compressed with DPCM technique while the remaining bands of coefficients are compressed with artificial neural networks. Metrics obtained: Peak Signal to Noise Ratio (PSNR) Mean Square Error (MSE) and Compression Ratio (CR) are tabulated for comparative analysis.