Image Retrieval Based On Binary Codes

With the development of the science and technology today,image retrieval hasbecome a hot topic for people to study and pay a lot attention to it.With the widespread of mobile devices as well as the gradual development of network technologyterminal,cloud storage technology is also under rapid development in recentyears.The growing maturity of these technologies cause the number of the images onthe internet have a exponentially growth.With this growth,the image retrievalbecomes more difficult.In the past few years,the traditional image retrieval technology based onhigh-dimensional floating-point feature descriptor.During the image retrieval,weneed to calculate the euclidean distance between these floating-point,but it is a bigchallenge to the huge images data,it cost a lot in computation,and the computingresources requirements are very strict,so it is impossible for the mobile terminal toachieve.To solve the problem above,people start to study how to change thehigh-dimensional floating-point to the low-dimensional binary featuredescriptor.The binary feature descriptor has a greater advantage in image retrieval,Itoccupies a smaller storage space,and it only has to compute the hamming distancebetween the feature descriptors during the image retrieval.It is quite simple,and theCPU performance requirements are not high.In response to this request,we provide a method to change the floating-points tothe binary encoding.First we map the high-dimensional floating-points data tolow-dimensional space with PCA. These data has certain characteristics in thelow-latitude space,they are symmetrically distribution of the axis of symmetry,whirling these data,mapping to a binary hypercube vertices,and encodingas different binary codes.During the changes,we use iterative method to make thebinary codes have the smallest similarity deviation.It is a greatly improve to theretrieval speed when we use the binary codes to do the image retrieval.Theexperiments after shows that our method is very useful,with the same length ofcode,our method has a great effect.