Temporal-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 temporal -frequency property of signal analysis and overcomes the shortcomings of traditional Fourier Transform which fails to present any temporal discrimination ability in frequency domain. Under Heisenberg’s uncertainty principle, it has been proved that it has optimal joint temporal-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 pattern detection related researches by means of Gabor Transform based joint Temporal-Frequency Analysis. The main contribution of the dissertation are as follows: 1) From the view of Temporal-Frequency Analysis,the temporal-frequency property of Gabor Transform is analyzed. Some Gabor Transform based typical applications are discussed, which include texture segmentation, image retrieval, object detection, and object recognition. 2) In edge detection, the edge output response based on Odd Gabor Transform is analyzed and a nonlinear adaptive threshold selection scheme based on Rayleigh distribution is proposed. Following the above works, the multi-scale analysis in Odd Gabor Transform domain is conducted and scale multiplication in Odd Gabor Transform domain for edge detection is put forward. The final experimental results show the proposed algorithm has stronger noise resisted capability and better visual effect compared with other common utilized edge detection operators. 3) Based on the rotation invariant property of Circular Gabor Transform,a robust object matching method by using weighted partial Hausdorff distance is proposed. In circular Gabor feature space with position information embedded, a coarse object matching is realized by using weighted partial Hausdorff distance. Then the final fine object matching is obtained by combining the circular Gabor features and the object’s shape information. The experimental results show that the proposed algorithm is robust to the influences from the cases of noise, occlusion, rotation variation, and scale variation. 4) For the consideration of computing efficiency of the above mentioned object matching method, a two-step object matching scheme based on hypothesis generation and verification is proposed. First the procedure of object matching is decomposed into multiple local optimizations to shorten the search paths for object candidates. Then the mean shift based local optimization technique with K-L divergence as similarity measure is utilized to generate fast the hypothesis set. In the process of local optimizing, a coefficient adjustment method is given to ensure the theoretic convergency of iterative optimizing.
修改评论