In this paper, we propose a rotation-invariant pattern-matching scheme for detecting objects in complex color images. The complexity and computational load for matching colored objects in arbitrary orientations are reduced significantly by the 1-D color ring-projection representation. It can rapidly select the possible locations of a reference template in the input scene by computing the normalized correlation of I-D color ring-projection patterns. Objects in the candidate locations are then verified by the pixel-to-pixel template matching. To make the pixel-based matching invariant to rotation. a color feature is used as pixel density, and the axis of least second moment is employed to estimate the rotational angle of a colored pattern. The proposed method has shown promising result based on experiments on a variety of natural and industrial images. (C) 2001 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved. [References: 19]
展开▼