The Achievement Of Image Search Engine Based On Sogou Image Dataset

With the rapid development of information technology, the text data is no longer the main source of information in people’s daily lives. Images, audio, video and other various types of multimedia information are growing explosively, so that the world becomes colorful. However, facing with the massive data, we encounter an urgent problem in what way to find the information quickly and accurately of multimedia field, then we witnessed the emergence of information retrieval which relies on the application of multimedia comprehensive analysis. The application of digital images in people’s lives becoming increasingly frequency, we searching needed images from a massive collection of images quickly and accurately, just depending on text annotations to achieve the goal appeared to be inadequate. The purpose of content-based image retrieval technology is how to effectively solve the problems of retrieve relevant images from an image database. The technology has become the focus of research at home and abroad, and is widely used in many fields now.In this thesis, based on an image dataset of Sogou, I establish a simple CBIR system of web mode to retrieve the similar pictures. Through a lot of experiments we retrieved the similar images by the features of images. The content and contributions of this thesis as follows:Firstly, we summarize the development process of content-based image retrieval, analyze the focus and directions of research in recent years, and introduce the value of image retrieval technology in various industries. We also describe a number of typical retrieval systems during the development process of CBIR technology. We introduce and analyze the related knowledge about image retrieval technology,Secondly, This thesis describes the image data set used in Sogou search engine, including the major categories of image datasets and the simple preprocessing about the dataset. It describes the framework of search systems, including interfaces of the system which is achieved by JSP, Servlet, CSS, etc. including uploading images, image preview, search results returned, etc. We use the server for feature extraction and feature similarity matching process.Thirdly, we introduce the related work of image retrieve based on global features. We adopt the color and texture features as image descriptors. We describe and analyze the algorithms of color features:color histogram, color moments, etc. Then we conduct experiments comparing the performance of different algorithms. We describe the algorithm which describes texture features, then adopt the uniform LBP texture features as the descriptor; Since we choose the Multidimensional vectors to represent global features, so we adopt the Euclidean distance to measure the similarity of color and texture features; We completed the retrieve experiments based on color and texture features and further analysis of the search results.Finally, we analyze local image features in the mariner described-SIFT algorithm, and explain in detail the steps SIFT feature extraction process, conduct SIFT feature extraction experiments; Due to the large dimension of SIFT features for the purpose of speeding up the efficiency, so we use BoW algorithm to create the index. We completed image retrieval experiment based on SIFT, and analysis and explain the search results of in-depth. We analysis and compare the effectiveness and performance based on global features and SIFT features retrieval results. We categorize the image datasets to improve image retrieval results.