| As the Internet technology is more and more widely used,there is more and more image data transmission via Internet.Not only the social network but also search engine will produce a large amount of image data.It is very difficult for people to look up the information by their naked eyes from this huge amounts of image data information.The performance of image retrieval system based on stand-alone computer could not reach the requirements of huge amounts of image processing.With strong capability of huge data processing,distributed computing platform is being widely used.Therefore,image retrieval system based on Hadoop is proposed in this thesis.The SIFT feature of image is very important for image retrieval system,which contains much important information of image.SIFT feature of image can be used to represent an image,so as to realize the retrieval of image retrieval system.In this thesis,SIFT algorithm is studied and improved.A more accurate way of gradient calculation based on the original SIFT algorithm is put forward,meanwhile,reducing the dimension of SIFT descriptor to 64 dimension by DCT and adding the NMI feature to SIFT descriptor for increasing the information are both proposed.In order to achieve the distributed processing of huge amounts of image data,HIPI is used to make the image data distributed stored.Program on MapReduce is wrote to extract the image features distributed,and classify the image features by K-means.An HBase table schema is designed to store the information of image data.In order to improve the efficiency of image feature matching,k-d tree is designed for storing the feature indexes.In the end,an image retrieval system based on Hadoop is designed and a case of the system implementation is presented.Compared to image retrieval system based on stand-alone computer,distributed image retrieval system is more quickly and which improves the efficiency of huge amounts of image retrieval and makes people more quickly to search the information they want. |
PDF Full Text Request