Camera Zoom Detection and Classification Based on Application of Histogram Intersection and Kullback Leibler Divergence

Abstract

This paper presents a novel compressed domain technique for detecting zooming camera in video sequences and its further classification into zoom-in camera and zoomout camera. The inter-frame block motion vector field serves as the input to the proposed system which is partitioned into four representative quadrants for analysis purposes.The histograms of these four quadrants are analyzed utilizing histogram intersection feature for zoom motion detection while the cumulative histogram of these four quadrants are analyzed utilizing Kullback-Leibler divergence feature for zoom motion classification purposes. Experimental validation carried out utilizing block motion vectors extracted using Exhaustive Search Motion Estimation algorithm as well as H.264 decoded block motion vectors demonstrate superior performance in comparison to existing techniques.