Theory And Application Of Adaptive Lifting Wavelet

With good localization and multiresolution property in space and frequency domain, wavelet is commonly known as "mathematical microscope" in signal analysis field. In recent years, wavelet analysis theory got vigorous development and were applied extensively in signal processing, pattern recognition, image processing etc. Our article combines the theory of wavelet analysis with palmprint authentication and plant leaf recognition. The main work and progress are as follows:1Traditional wavelet transform can’t adapt the detail and edge characteristics of images when applied to digital image processing, so it protects weakly to these features. It is hot research topic that how to further improve the image protection ability to the detail and edge informations on the basis of wavelet analysis theory and the image’s local characteristics. In this paper, we proposed an adaptive lifting wavelet(ALW) structure successfully, which is applyed to palmprint identification and plant leaf recognition. Experimental result shows the effect is remarkable. Based on pixel correlation, adaptive lifting wavelet can take different operations according to the image characteristics, i.e. for smooth regions, we compute the weighted average of the central pixel and its eight horizontal, vertical and diagonal neighbors, while don’t perform any filtering for less homogeneous regions. It can effectively protect the detail informations such as the discontinuity and singular point, which is critical step in image feature extraction and image processing.2In conjunction with pulse-coupled neural network(PCNN), adaptive lifting wavelet is employed in palmprint identification. The palmprint image should be firstly preprocessed so that we can capture the same size interested region. Secondly we construct an adaptive lifting wavelet scheme to decompose a preprocessed palmprint image into several subbands. Then the pulse-coupled neural network is employed to decompose each subband into a series of binary images, the entropies of which are calculated and regarded as features. Thirdly, a support vector machine-based classifier is utilized in the classification stage. Experimental results show that the proposed approach yields a better performance in terms of the correct classification percentages compared with the recent on-line palmprint recognition algorithms. It is also shown that the proposed approach yields observably low computational cost and can be easily implemented with hardware.3A new dual-scale algorithm based on adaptive lifting scheme is proposed, which is applyed into feature extraction of plant leaf recognition. Firstly, the leaf images from different plants species are preprocessed into the same size. Then, the preprocessed plant leaf images are decomposed by adaptive lifting scheme three times. The third level subbands are filtered by the wide gaussian function, while the second by wider and the first by widest. After applying center-symmetric local binary pattern (CS-LBP), the corresponding fuzzy entropies are then calculated and regarded as features so that the important characteristics can be emphasized. Finally, SVM-based classifier is employed to plant recognition. The experimental results demonstrated the strong robustness of our algorithm for plant leaf recognition.