Study Of Hashing Algorithm For Large-scale Image Retrieval

With the rapid and continuous development of the multimedia technology,network technology and popularization of the digital imaging equipment,the database containing a large number of images continue to occur,application of large—scale image data is facing a huge challenge.Image retrieval technology plays an important role in machine learning,artificial intelligence and computer vision and so on.Content based image retrieval technology extracts the feature descriptors of images and calculates the similarity of the image features to retrieve similar images of the query image.Nowadays,with the explosive growth of data and the increase in the dimension of the data,the large amount of data storage and calculation in large scale images brings new problems to the image retrieval technology.Recently,the vision community has devoted a lot of attention to the problem of learning hash binary codes for large-scale image processing because of its low storage cost and fast retrieval speed.In image hashing method,learning hashing functions to embed high-dimensional feature to Hamming space is a key step for accuracy retrieval.In order to obtain an effective hash function and improve the efficiency of image retrieval,the following research work has been carried out in this paper:(1)PC A(principal component analysis)technical has widely used in compact hashing methods,and most these hashing methods adopt PCA projection functions to project the original data into several dimensions with real values,and then each of these projected dimensions is quantized into one bit.To avoid the real-valued project with large variances and quantization error,in this paper we proposed to use Cosine similarity projection for each dimensions,the angle projection can keep the original structure and more compact with the Cosine-valued.The experimental results on public image dataset show that the proposed algorithm is superior to other hashing algorithms.(2)And most existing approaches using a single global feature to learning binary codes,to overcome the disadvantage of global feature that lose local information needed for image retrieval tasks.In this paper we propose a new method that combined global and local feature to learning compact binary codes.In our framework,we first extract the HOG(Histogram of Oriented Gradient)feature and the GIST feature from each images,the second step is to obtain the binary code by fusion the two feature vectors.Finally,the experimental results are compared with those of the same hash algorithm using single feature.The results show that the multi feature method proposed in this paper can improve the retrieval accuracy.