Image Retrieving Based On SIFT Feature And K-means Cluster Method

With the development of the internet and multimedia, the sharing of data becomesmore and more important, the issue of image retrieving has become the core problem.Meanwhile, the performance of mobile devices is rapidly improved, high-definition camera,high-frequency processors, multi-core technology are already mature embedded into mobiledevices, these hardware devices provide a solid foundation for mobile image retrieval in theend. At the same time, cloud storage technology has matured, Internet terminal receives theimages from every corner of the world hundreds of millions of daily, data capacityexponentially rapid expansion. On this basis, it has proposed the use of the mobile deviceside image retrieval techniques search cloud storage side picture. Real-time image retrievalend mobile devices irreplaceable significance, and now many commercial mobile devicesend real-time image retrieval applications have been widely used.Retrieve image in such database is challenge, in order to accelerate the retrievingspeed, the problem of the dimensional of image should be concerned. A lot of retrievingalgorithm can be employed in low dimension space but not in high one. Because of thisscene, in computer vision, in the area of image retrieving, the hash coding of the data hasbecome interested by more and more people.In computer vision there has been increasing interest in learning hashing codes whosehamming distance approximates the data similarity. The hash function is very important inboth quantizing the vector space and generating similarity-preserving codes. In our paper,one hash method has been introduced,this hash method is proposed based on the classiccluster algorithm the k-means cluster. We proposed the hash method based this brilliantcluster algorithm, this hash method can complete two task, one is clustering the origin data,then the origin data can be clustered into many different cell, then the hash method willlearn the binary code. Different cluster center will have different binary code, the binarycode will be used in the following stage of the retrieve algorithm, the retrieve speed will beaccelerated thanks to the binary code. Our proposed method is look-up based method. Thismethod quantize the space into cell via k-means. k-means is a more adaptive quantizationmethod than using hyper planes, and is optimal in the sense of minimizing quantizationerror. Our proposed hash method can reduce the quantization error by using k-means-basedquantization for a larger number of bits. A large number of experimental results show that our proposed method is better than alot of state-of-art hashing encoding method.