Font Size: a A A

Research On Content Based Image Retrieval

Posted on:2003-10-25Degree:DoctorType:Dissertation
Country:ChinaCandidate:X H SunFull Text:PDF
GTID:1118360062476512Subject:Pattern Recognition and Intelligent Systems
Abstract/Summary:PDF Full Text Request
Content Based Multimedia Retrieval is a retrieval technology based on the media features, such as the color, texture and shape of the image, the scene and movement of the video, and the tone, highness and tamber of the audio. Content Based Multimedia Retrieval includes the image, audio and video retrieval, among which this paper does the most research on the image retrieval. Content Based Image Retrieval (CBIR) means retrieving images based on the image content description features directly. In the past ten years, CBIR is one of the most active hot points in the current research fields such as the computer vision, the image database and the knowledge mining and so on.It is natural that the image under observation is divided into the object and background parts by the human vision. The image retrieval performance is evaluated mainly with the object-based measurement. Generally speaking, the object regions should be the most attractive places of the image, with the sharpest contours, the fullest edges and the most prominent areas. In the image retrieval the object regions accord with the human vision perception better than the background region, and the image features extracted from the object regions meet the needs of the image retrieval better. This paper does research on the color image retrieval based on the object regions, and presents the corresponding retrieval algorithm. Experiments show that the color image retrieval algorithm based on the object regions is superior to that based on the global image, and that the color image retrieval result based on the object regions accords with the human vision perception better.Texture is an important image feature that is hard to describe and has no acknowledged precise definition yet. Texture, as an important component of the human vision, can reflect the depth and surface information, and supply the human vision with the recognition and understanding information. Coarseness, contrast and directionality, according to the human perceptive experience, are three of the most principal features for the texture discrimination, and coarseness is the most essential and important textural feature. This paper analyses the previous Rosenfeld coarseness algorithm and improves it on the selection of neighborhood sizes and the calculation of neighborhood average differences, and presents the improved coarseness algorithm. Experiments show that the improve coarseness has higher texture discriminability and better rotation invariance, and that the image retrieval result based on the improved coarseness is superior to that based on the previous coarseness.Trademark plays an important role in the market economy, reflecting themerchandise quality and the manufacturer credit standing. The trademark image is a kind of artificial image without the complicated objects and background of the nature image. Each part of the trademark image has the obvious division to the other part and forms die independent impression in the human brain relatively. Extract several disconnected subimages from the trademark image, and the subimage features can reflect the local image feature that supplements the global image feature each other in expressing the image content hi the trademark image retrieval the subimage features can be fused to achieve the better performance. This paper puts forward the trademark image retrieval algorithm based on the subimage feature fusion, and does research on the subimage feature fusion criterions. Experiments show that the trademark image retrieval algorithm based on the subimage feature fusion is superior to that based on the global image feature, and that the subimage feature fusion weighted criterions are better than the normal criterions, among which the minimum weighted average criterion is best.This paper designs a CBIR system as the test bed for retrieval algorithms, which is an experimental frame system. The system platform for developing this CBIR system is Windows 2000, and the development environment is Visual C++ 6.0. Further work will be done t...
Keywords/Search Tags:Content Based Image Retrieval, Object Regions, Textural Coarseness, Subimage Feature Fusion, Retrieval Precision, Retrieval Recall
PDF Full Text Request
Related items