基于视觉内容的图像分类与检索计算机科学与技术专业论文.docxVIP

哈尔滨 哈尔滨工业大学工学硕士学位论文 哈尔滨 哈尔滨工业大学工学硕士学位论文 - - PAGE IV - - - PAGE III - Abstr act With the vigorous development of multimedia technology and internet, image has become an important information carrier. It is very hard for users to find their real needs in the growing mass of image data, so the technology of image categorization and retrieval gets multimedia researcher’s attention. In our paper, we investigate the image categorization and retrieval from image visual feature extraction, the construction of multi-classifier and the relevance feedback for image retrieval, and we implement image categorization and retrieval system on the basis of our research work. In the extraction of image visual feature, we have researched the popular methods of image visual feature extraction from color, texture and shape; we also analyzed the advantages and disadvantages of every method. In the construction of multi-classifier, basing on the research of SVM multi- classifier construction, we proposed an adaptive feature-weight adjusted categorization algorithm. When we use a number of features to express the image content information, our categorization algorithm can auto-matically learn each feature’s weight. The experiment result showed that our scheme improved the performance of image categorization comparing with general approach. Furthermore, when our scheme trains and categorizes different types of images, it has a certain adaptive capacity and adjusted ability by automatically learning each feature’s weight. In the relevance feedback for image retrieval, we firstly analyze the short- comings of SVM for learning of relevance feedback and study the solutions proposed by other researchers, then we propose a relevance feedback algorithm by combining query expansion and the learning of relevance feedback based on SVM. The experiment result showed that our scheme was better than the method simply based on SVM and simply based on query expansion. Keywor ds image categorization; i