Performance Analysis of Neural Network Architecture Combined with DWT for Image Compression

Abstract

Neural networks are significantly used in signal and image processing techniques for pattern recognition and template matching. In this work neural networks are used for image compression. In order to improve the performances image compression algorithm, DWT is combined with NN for achieving better MSE and increase in compression ration greater than 100%. NN architecture achieves maximum of 98% with use of four neurons in the hidden layer, with selection of LL sub band only the compression is improved by another 75%. The proposed architecture is analyzed for 20 images and MSE is found to be improved by a factor of 20%. Daubechies wavelet filter and Haar wavelet filters are used for DWT, input layer with one hidden layer and output layer consisting of tansig and purelin function us used for compression. The design proposed is suitable for high resolution image compression.