Research On Image Classification Based On Region Of Interest Detection

With the rapid development of Internet and multimedia technology,more and more intelligent terminal is widespread and there are massive image and media information on the Internet.Thus,how to do the images classification work in a effective way is both important to administrators and users,and this work will be promising.Now,images classification draws more and more scholars' attention to it,and become a hotspot issues in computer visual domain.Based on the analysis of relevant research about image classification at home and abroad,combined with traditional and mainstream object detection and image classification methods,two kind of new methods which based on region of interest detection for image classification were proposed.Distributed platforms are used to speed up some procedures of proposed method.The main works of this thesis are as follows:(1)In the process of image classification,the image region containing object which plays a decisive role is indefinite in both position and scale,and it could not get a high accuracy of image classification by using Spatial Pyramid Matching(SPM)directly.Therefore,a method for improving the performance of image classification based on Region of Interest Detection and Spatial Pyramid Matching is proposed.After extracting Region of Interest(ROI)and background region,Spatial Pyramid was used to multiscale partition the image and describe the feature with bag-of-words,and merge its histogram features to train a classifier to get a initial result.At last,rescoring the initial result with ROI confidence value and get the final result.Experiment shows that this method could improve the classification accuracy.(2)Since just a single feature and classifier are used to image classification in the above method,and it has some limitations.A method called Multi-feature fusion based on region of interest detection is proposed to combined with some kinds of features and classifiers.Experiment shows that this method could improve the classification accuracy even further.(3)Since the single machine could not solve the problem timely in the procedure oftraditional image processing,a new frame which is used for image processing based on distributed platforms is proposed,and could make the best of efficiency and extensibility of distributed platforms.Experiment shows that parallel processing could reach a good Speedup ratio and improve the efficiency of image classification.