| Due to the steady growth of computer, multimedia, and Internet techniques, a huge amount of images are available. Currently, rapid and effective searching for desired images from large-scale image databases becomes an important and challenging research topic. Content-based image retrieval (CBIR) is the set of techniques to address the problem of retrieving relevant images from an image database based on automatically derived image features. In recent years, CBIR is a very active research direction and has been applied to many fields.In this dissertation, lots of exploratory research work has been done around some key techniques of CBIR, which include low-level feature extraction, similarity measure, relevance feedback and so on. The presented study is the current research focus of image processing and information retrieval. Thus, its research has both the theory and the application value.The main contributions of this dissertation are summarized as follows:Firstly, several key techniques and algorithms of CBIR are deeply analyzed and discussed, such as, the relevance feedback, the low-level feature descriptions including color, shape, and texture, and the similarity measure between images. Moreover, by some experiments tested under the same conditions, we report the comparison results of many classical methods.Secondly, an image retrieval approach based on salient edges is proposed. It introduces an independent edge self-reinforcement algorithm to link edge curves and strengthen the salient edges. According to the reinforced results, the salient edges of the image could be easily extracted, and the image features are thus described by those salient edges. Then, a "many-to-many" matching strategy is used to measure the similarity between two images. It allows a salient edge of one image to match with several salient edges of the other image, and the final similarity between them is determined by all the valid matching. The matching principle can be tolerant to inaccurate edge extraction.Thirdly, a salient interest point based image retrieval algorithm is presented. This algorithm firstly uses an adaptive filter to process the image. Then, the popular way is adopted to detect the interest points. The extracted interest points are always located in the salient edges since the adaptive filter has the ability to smooth the details and enhance the salient edges of the image. Afterwards, the image is represented by the color feature of small areas centered at extracted interest points. This feature combines the shape feature implied by interest points with color feature extracted from those small areas, which decides the similarity between two images. Many experimental results show that this approach is simple and effective.AbstractNext, a novel image retrieval method based on attention objects in the image is reported. One of its important advantages is that, it accords with the human's subjective concepts while querying images. At first, it analyses the image content and acquire the attention objects. Then, images are matched and indexing is performed based on the features of those extracted attention objects. As the core technique of the method, the attention object extraction is implemented by incorporating a visual attention model with the seeded region growing that is a famous technique in the image segmentation. Lots of experiments demonstrate the retrieval method is very effective. Additionally, the technique of attention object extraction could be applied to other research areas such as, content-based image compressing, target recognition, and so on.Fifthly, a texture description and accessing algorithm based on the Variogram function is introduced. It has the ability to describe various kinds of texture, that is regular texture and irregular texture, and further search for images by texture features within a unified framework. Before retrieving images, this method is able to automatically predict the texture type of the query image, and then employs different schemes to deal with diffe...
|
PDF Full Text Request