| Summary: | 博士 === 國立臺灣科技大學 === 電機工程系 === 103 === This dissertation presents some techniques on image feature excavating and image characteristic modeling for several applications. The image feature can be simply constructed and generated by considering the underlying statistical property of an image as well as image content and characteristic. Based on this premise, four various research topics are included: 1) Singular Value Decomposition (SVD)-based image watermarking, 2) statistical-based vehicle verification, 3) image retrieval using halftoning-based Block Truncation Coding (BTC), and 4) halftoning-based BTC image restoration. Among these, the vehicle verification employs the statistical modeling of an image to generate an image feature descriptor, and the other methods extract the image feature descriptor by considering the image content and characteristics.
Some security attacks and ambiguity issues on the SVD-based image watermarking are explored and discussed in this dissertation. These attacks are delivered with the demonstration on the former published SVD-based image watermarking scheme, and the solutions are provided in this dissertation. Some methods with the statistical modeling on feature descriptor generation are also presented in this dissertation for the vehicle verification task. An effective approach is presented in this dissertation to generate the image descriptor from the halftoning-based BTC compressed data stream for image retrieval and classification. An additional approach on image restoration of halftoning-based BTC is also presented in this dissertation. The performance of the proposed feature descriptors are extensively investigated and tested for their suitability on the image watermarking, vehicle verification, image retrieval, and image restoration applications. As documented in the experimental results. The proposed methods can be effectively applied to these domains, and thus they can be very competitive candidates to the practical usages.
|
|---|
