Intelligent Object Detection Using Trees

摘要

In this paper a method is proposed for detection and localisation of objects in images using connected operators. Existing methods typically use a moving window to detect objects, which means that an image needs to be scanned at each pixel location for each possible scale and orientation of the object of interest which makes such methods computationally expensive. Some of those methods have made some improvements in computational efficiency but they still rely on a moving window. Use of connected operators for efficiently detecting objects has typically been limited to objects consisting of a single connected region (either based on simple or more generalized connectivities). The proposed method uses component trees to efficiently detect and locate objects in an image. These objects can consist of many segments that are not necessarily connected. The computational efficiency of the connected operators is maintained as objects of interest of all scales and orientations are detected using two component trees constructed from the input image without using any moving window.