Design And Implementation Of Image Search Based On Locality Feature Detection

There are many similar or partial-duplicate images in the Internet. With technologies of content-based image retrieval, people can find duplicate versions of a picture which they care through Web and use them for deep processing, copyright protection and so on. This thesis focus on the locality feature image retrieval theories and approaches and discuss how to design and implement an Internet image search engine in order to make useful search results and verify some state-of-art image retrieval algorithms.The most successful approach in related studies is the Inverted Index model with BoVW(Bag of Visual Words), so we dig into the model and do some further research. Firstly,we build inverted index with Vector Quantization to accelerate the image searching. Secondly, we use spatial-coding to filter false matches during the stage of indexing and searching. With the combination of these theory and methods, we build a useful image search engine, and give a feasible solution which attains the time and space efficiency of large-amount images search.The last part of this thesis describes how the Internet image search engine was built in detail. The system contains web-image crawler, image file storage subsystem and Index and Search subsystem,which implements the algorithms we discussed in chap three and chap four. The search result of our search engine and the experimental outcome compared with similar system show that the software was working well and valuable.