الفهرس | Only 14 pages are availabe for public view |
Abstract In this thesis, a technique for automatic selection of best mother wavelet that is suited for 1D signal compression is proposed and implemented. The proposed system is a pattern recognitionbased solution that is based on extracting some features of the signal to be compressed then a classifier is used to select the best wavelet for this signal under test. The feature extraction block uses a modified fractal analysis approach for solving the problem under study. A neural network is trained by the fractal analysis features to estimate the index of the best wavelet required to compress the signal under consideration. During the study, a training set of different categories of 1D signals is used during the learning phase of the classifier. After training, the neural networkbased classifier chooses automatically the best wavelet function for compressing a given signal. Practical study for compressing a set of 52 1D signals has been carried out to build a knowledge base for the neural network classifier. According to this step, four mother wavelets have been proved to be suitable for compressing more than 92% of the ?onedimension signals? under test in case of High Quality signal to be reconstructed. |