The Research Of Image Retrieval Based On Embedded System

With the development of Internet and digital technology, the content of image database is constantly enriched and improved. The problems users facing are how to search the satisfactory information from the large amounts of image information source quickly and accurately instead of the lack of information. Therefore, scholars have put forward the content-based image retrieval (CBIR) technology. Furthermore, with the continuous improvement of embedded equipment performance and technology, the embedded image retrieval system becomes a hotspot at home and abroad. Establishing an effective image description and retrieval mechanism, based on various embedded platforms, has become one of the key problems which need to be resolved urgently. This paper will regard this as the research subject to build a system platform, refine the image feature extraction algorithm, and enhance the image retrieval performance.This paper constructes a content-based embedded image retrieval system in the client-server mode. The client which is the mobile phone constructed with the carrier of the Android system can implement the collection of images and send them to the server. The server takes desktop computer as the carrier. When the server receives the images which sent by the client in the personal computer, it will retrieve the images according to the rules and give some feedback to the client. This thesis details the construction method of the server and client.On the aspect of the image feature extraction, the extraction methods of low level features, like color, shape, texture and spatial relation, are analyzed. The gap between low-level features and high-level features like semantic feature are also studied. Moreover, the local feature extraction method is also mentioned. In order to enhance the accuracy of content-based image retrieval and speed up the retrieval time, the paper retrieves the images with local characteristics by adopting and tests the accuracy of the retrieval system based on the bag-of-words model. What’s more, this paper designed a contrast experiment which compared the algorithm proposed in the paper with the retrieval algorithm based on single image characteristics. The final results show that the improved image retrieval method enhances the retrieval accuracy at some level and has a theoretical and practical significance.