Research On Content-based Image Retrieval And Practice

With the rapid development of information storage technology and the spread of the Internet, a large capacity image databases that contain different contents in the images are generated. As a result, it has become an increasingly important topic how to effectively manage and retrieve these images. Especially when the current image databases contain rich contents in the images, their features cannot be adequately expressed with the traditional text-based methods. The Content Based Image Retrieval is a new approach based on Machine Vision, Pattern Recognition which performs the computer-centered image retrieval according to the content of images. It greatly reduces the burden of manual annotation, improves the image retrieval rate and provides a broader prospect for the image retrieval.On the basis of some key technology and theories of content-based image retrieval, this thesis has conducted the following research: 1) A novel shape retrieval method based on the technology of the InnerDistance is proposed ,and is validated its effectiveness by experiments on two public datasets; 2) It has introduced the HOG feature based image Retrieval in the real Vehicle License Plate Character searches, since the technology of HOG algorithm can better describe the shape and outlines of images; 3) The KD-tree algorithm is introduced in the two methods mentioned above in order to accelerate the retrieval speed.The experimental results indicate that the above-mentioned methods for the content-based image retrieval have greatly improved the recall rate and precision rate than the traditional retrieval method. It has certain reference value and practical implications to promote the development of content-based image retrieval technology.