Research On Image Retrieval System With Automatic Annotation Algorithm

With the increasing popularity of digital cameras customers and widely spread of multimedia information in network, digital images appeared and accumulated quickly. How to manage and queries such large-scale image data effectively has become a key research project in the field of multimedia technology.Researches on image retrieval technology has begun since late1970s. Firstly it starts with Text-Based Image Retrieval which needs text labels added manually. However, as the origins of images and types of image formats become more and more diversity recently, it is not practical at all to label huge image data manually. Then in1990s, Content-based Image Retrieval technology showed up, which implements the query by using low-level visual features like colors, grain, etc. But, it was soon found that there are so-called Semantic Gap between the low-level features and high-level semantic. To narrow down the Semantic Gap, another technology showed up. It is a retrieval technology of semantic-based automatic label, which achieves the goal of high-level semantic retrieval by automatically labeling the image database.This essay puts forward a new kind of image label framework which bases on ontology. Combined with the relationship of semantic concepts in the fields, it gets more accurate concepts of image high-level semantic by using level probability filter and then realizes the semantic annotation of image. When first labeled, it extracts and classifies primitive images in the training collection, and establishes association probability between the primitive classes and semantic concepts using statistic methods. It also uses Bayesian algorithm to calculate the posterior probability between ontology and the concepts to be marked in the image, and then choose the larger posterior probability vocabulary to label the image. When labeled later, it obtains high-level semantic of the image after considering the semantic relationship between ontology and concepts, thus the automatic annotation of unmarked image semantics has accomplished. Secondly, it uses the platform collected by Java language, the integrated environment of MyEclipse, and MySql database and finishes the design and implementation a prototype system of image retrieval which based on HSV and RGB color features. This system extracts image features in the feature-extracting module by using external examples image query style, then searches the matches of the features in database and finally outputs the retrieved target image in the result display module.