Gabor Transform Based Feature Extraction And Its Applications

Time-Frequency analysis is one of most important tools for signal processing. Gabor Transform, which is also named by Short-Time Fourier Transform or Windowed Fourier Transform, shows the joint Time-Frequency property of signal analysis and overcomes the shortcomings of traditional Fourier Transform which fails to present any time discrimination ability in frequency domain. Under Heisenberg's Uncertainty Principle, it has been proved that it has optimal joint Time-Frequency resolution. Based on the studies on the physiology of mammal perception system, it has been shown that 2-D Gabor Elementary Function can fit well the receptive fields of the majority of simple cells in mammal visual cortex. All works presented in the dissertation are to conduct patern detection related researches by means of Gabor Transform based joint Time-Frequency Analysis.The main contribution of the dissertation is as follows:1) From the view of Time-Frequency analysis, the property of 2-D Gabor Transform is analyzed. We used its energy distribution in Time-Frequency domain to extract palmprint's feature and obtained better recognition rate. As another method, we used two-channel wavelet and three-channel wavelet to extract palmprint's low-prequency energy as our feature, used SVM to recogniaze, we obtained good recognition rate finally.2) Used Gabor Wavelet to retrieve texture images. We used Gabor Filter with several scales and directions to extract texture image's feature, and obtained good retrieve rate.The main work includes:Introduce the definition of complex Gabor Transform, including critical-sample and excessive-sample complex Gabor Transform. In order to compute fast, we used Real Discrete Gabor Transform instead of Complex Discrete Gabor Transform. We found that Complex Discrete Gabor Transform coefficient could be obtained by Real Discrete Gabor Transform coefficient by analyzing their relationship. We also introduce Gabor Wavelet and analyze its property, supplying theory basis.We look back the history of palmprint recognition. Palmprint recognition, as an important biometric method, one of the most important parts of it is the feature extraction. Here a new method was proposed to extract the palmprint energy distribution feature in Time-Frequency domain using Real Discrete Gabor Transform. Finally we made a feature match based on this. This is a simple and effective method. As another method, we used two-channel wavelet and three-channel wavelet to extract palmprint's low-frequency energy feature, also obtained good recognition rate.The last part is retrieval of texture image. First, we introduce its research history. Because of texture image's periodicity and direction, we use Gabor Wavelet with several scales and directions to extract image's periodicity and direction feature, select its equal and difference to compose feature vector, as a result, we obtain good experiment results.