Display Optimization For Image Browsing
In this paper, we propose a technique to visualize a multitude of images on a 2-D
screen based on their visual features (color, texture, structure, etc.). The resulting layout will
automatically display the mutual similarities of the viewed images. Furthermore, audio features, semantic
features, or any combination of the above can be used in such a visualization. The original high
dimensional feature space is projected on the 2-D screen based on Principle Component Analysis (PCA). PCA
has the desired property of being simple, fast and unique (i.e. repeatable) and the only linear
transformation that achieves maximum distance preservation in projecting to lower
dimensions. Furthermore, we have developed a novel technique for solving the problem of overlapping
(obscured) images shown in the proposed 2-D display. Given the original PCA-based visualization, a
constrained nonlinear optimization strategy is used to adjust the position and size of the images in
order to minimize overlap (maximize visibility) while maintaining fidelity to the original positions
which are indicative of mutual similarities. A significantly improved visualization of large image sets
is achieved when the proposed technique is applied.
|