The Analysis Of Image Similarity Based On Mapreduce Model

With the rapid expansion of information, the image which is an important form of bearing information and its retrieval technology has become an important research direction in recent years. As the keyword-based retrieval mode can’t meet the needs of users, the CBIR(Content-Based Image Retrieval) technology has been proposed. And the underlying visual feature extraction also made some progress, such as color, texture, shape feature extraction have got a variety of algorithms. Although texture and shape features not yet appeared relatively uniform extraction methods, many kinds of methods retained the visual characteristics of the image from different aspects. On one hand, the underlying visual feature extraction is becoming more and more refined. On the other hand, the rapid growth of the amount of data let the stand-alone CBIR system face with enormous challenges. Therefore the combination of distributed computing and CBIR system is imperative. This paper introduces the research from the following aspects:The study analyzed CBIR system from the applicability and extracted in color, texture, shape features:(1)In color feature, the study compared the extraction based on the RGB model and the HSV model, and the result shows that the advantage of the HSV model on retrieval effect.(2) In texture feature, the study used GLCM-based texture feature extraction, and used texture image and non-texture image for testing, the result shows that the texture features is not general.(3) In shape feature, the study extracted based on the edge direction histogram, after testing with several image, we can find the retrieval rank is good, but the similarity value is not easy to distinguish. Finally, integrating color and shape features, the results can be more comprehensive reaction to the impact of underlying visual features.Considering the stand-alone CBIR system will spend a lot of time, distributed computing to solve this problem is used. In order to use distributed computing, the Hadoop system and MapReduce programming model is chosen. So the work is focused on the design of image matching module based on the MapReduce architecture, so the match modules can take advantage of distributed computing. Finally, the advantage of this architecture is verified by experiment.