Visualized Exploratory Search Of Massive Images Based On Correlation

Nowadays, with the development of network technology and widespread use ofdigital equipments, the image data has posed an explosive growth, the method thatusing a single computer to process images has been unable to cope with the real needs.Meanwhile, the current image searchers have formed the habits to enter keywordswhich represent the information of interests, then they submit these words and receivea ranked list of images. These search engines are designed for "one-shot search",when a query does not hit, users need to redesign the query, so the search engines canreturn satisfied results only if the users have clear destination. However, when usersdo not know how to reach the target, or the target itself is ambiguous, how to organizeand use the image data, how to find out the wanted images from the mass images andhow to make users to understand lots of images from the returned lists and get therelationship between images become a hot and difficult problem.The research field Exploratory Search embedded in the field Human ComputerInteraction, aims to come to a next generation search interfaces beyond documentoriented such as Google, Bing and so on. Exploratory search are driven by curiosity ora desire to learn about or investigate something, which emphasis on interaction withthe users. The main purpose of visualization is to make data visualization, makingpeople identify and browse information quickly. Mobile devices like smart phoneswith powerful functions can interact with people freely, and it is true that more andmore people own these devices and they are trend to search information by thesedevices. Thus, study on a mobile device for image visualization exploratory searchhas a considerable importance.In order to solve these problems, we have made the three following works:(1) In order to solve the existed stand-alone image processing problems, such asinsufficient memory and long processing time, we use the Hadoop platform to processmassive images, and then use the spectral clustering methods to get the "bag of visualwords "(BOVW), making the abstract image information be "visual vocabulary"。(2) We first proposes a method using the analysis of co-occurrence technology todefine the correlation between images, which can solve the problem of understandingthe relation between images difficultly.(3) We implement a visualization exploratory search system of massive imagesbased on correlation, which can solve the problem of un-convenient input to androidsmart-phone users, increasing the friendliness to these users.Compared with the existed search systems, the proposed search system in thispaper has three advantages as followed:1. Support the user to interact with the systemin a graphical interface.2. Enable users to distinguish the relationship between thequery-image and return image more clearly.3. Induce users to go to the next step ofexploration, getting richer and more interested information. Experimental resultshows that the proposed method can effectively improve the ability of users to accessrelevant images and get the relationship between images.