Classified Image Compression using Wavelets and Neural Networks

In this research it is proposed to adaptively select the best wavelet based on statistical features viz Image activity measure (IAM), Spatial Frequency (SF) and randomness feature viz Entropy filtered features, standard filtered features (STD) and the performance of the selected wavelet is measured in terms of Transform Coding Gain (TCG), Mean Square Error (MSE), Peak Signal to Noise Ratio (PSNR) and Compression Ratio (CR). Based on these selected features the images are classified using Counter propagation NNs and SVMs. Biorthogonal wavelet is most suitable for compression of natural images, symlet wavelet is for SAR images and Daubechies wavelet for medical and cartoon images. SVM has resulted better recognition accuracy compared to Counter Propagation Neural Networks in image classification.