A Hybrid Bi-Orthogonal Wavelets, Neural Networks and DPCM based Image Compression Approach for Performance Analysis

In view of the massive parallel architecture and generalization ability of neural networks to remember the inputs for untrained data, the computational simplicity of wavelets, ability of Differential Pulse Code Modulation (DPCM) to reduce the redundancy in the information, an effective hybrid image compression system combining the advantages of wavelets, artificial neural networks and DPCM is implemented. Quantization and Huffman encoding schemes are also used to compress the input image. Here the low frequency coefficients are compressed with DPCM technique and the high frequency coefficients are compressed with multiple feed forward neural networks. Objective fidelity measures like Peak Signal to Noise Ratio (PSNR) Mean Square Error (MSE) and Compression Ratio (CR) can be obtained for performance analysis. Wavelet transform reduces the blocking artifacts’ of cosine transform while the neural networks minimize the Mean Square Error (MSE).Empirical analysis and metrics calculation is tabulated for the purpose of relative analysis.