Cubic Style Browsing System for Large-Scale Image Database

This paper presents a cubic style browsing scheme for large-scale image datasets. In our method, we map all images into six 2D planes spanned by the axes defined by low-level features such as color, edge, image stability, centroid coordinates, maximum object size, and roundness. We form a cubic by combining these planes for image browsing. When a user selects one image, its neighbor database images on the same plane to which the selected image belong then are displayed for the user hierarchically. Experimental results on CIFAR-10 show that the effectiveness of our method is better than that of a simple display approach such as principal component analysis.