A Primitive-Based Multistage Image Retrieval System

With the development of the multimedia and network technology, the application of the image is extensive and the content-based image retrieval technique has already become the studied focus. It has combined the technologies, such as image processing, pattern-recognition, vision understanding, artificial intelligence, computer vision, databases, etc. It is an extremely promising new technology in research and application.The text-based image retrieval and content-based image retrieval technique are two kinds of main methods to image retrieval at present. The text-based image retrieval technique utilizes the manual image annotation as image character for retrieval and is a precise matching. The content-based image retrieval utilizes computer to automatically extract image content (like color, texture, shape, object and space-relation, etc) as image character and is a similarity measure. The text-based image retrieval can be traced back to the late 1970's and its technology is already mature. But the content-based image retrieval what proposed in the early 90's and in process of speedinessdevelopment is used to solve a series of problems brought by manual image annotation in the text-based image retrieval.In this paper we narrate the developing history of image retrieval technique and current situation, especially introduce all respects of the content-based image retrieval technique, deeply analyse existing problems and each advantage or defect in two kinds of image retrieval technique and have proposed a primitive-based multistage image retrieval system on the basis. Through build the "semantic dictionary" module and the " primitive-feature base" module in retrieval system, have finished the organic integration of two kinds of retrieval methods. In this system, we set up the relationship between vision character of ground floor of image and semantic character on the senior level of image, and primarily solve "semantic gap" between the simple visual characters and the abundant semantics delivered by an image. On the existing technical level, it can meet the need of people about semantic image retrieval and it is a benefic attempt to semantic image retrieval technique.