Applying the Generalized Compass Operator for Image Dominant Orientation Estimation

Abstract

Evaluation of the orientation of objects on frames from digital cameras is one of the problems of intelligent image analysis and processing. In this paper, we focus on the two-stage algorithm for image dominant orientation using first rough estimation of orientation angle through principal component analysis of gradient covariance matrix and second more accurate orientation estimation using the generalized compass operator. We compared two types masks of compass operator – Prewitt-like and extended Prewitt-like – for orientation estimation task. We demonstrated that for extended Prewitt-like mask with a size of 5?5 pixels using the generalized compass operator provides an estimation error of the dominant direction no more than 1° compared with methods, based on principal component analysis technique.