SCENE COMPLEXITY ANALYSIS USING RANDOM CURVES

摘要

We present a novel method for the analysis of scene complexity based on integral geometry. We use random families of ellipses and measure the number of intersections between these curves and the objects in images. The intersection number gives a good measure of the scene complexity - the higher the complexity of the image the higher would be the intersection number. The proposed method is applicable to arbitrary scenes and to any types of grids for digitization. Our method improves an earlier work in which random lines were used to estimate scene complexity. For objects that are not of square shapes, a random family of ellipses is found to be better than a random family of lines in measuring the scene complexity. The numbers of intersections correlate to the density of the curves. Also, the orientation of the family of ellipses influences the number of intersections to some degree. Experimental results are given to illustrate the proposed method.