Estimating the Dominant Orientation of an Object Using Image Segmentation and Principal Component Analysis

An object's orientation can often be a hurdle in computer vision applications. Assuming the object has a major axis, i.e., is longer in one of its dimensions than in others, the object's dominant orientation can be found. Knowing and compensating for an object's orientation may simplify processes such as recognition, segmentation, template matching, etc. However, solving this problem with no prior knowledge of the object's properties is not trivial. A solution is proposed which uses an image segmentation process that requires minimal prior information of the object, followed by feature extraction, and finally principal component analysis. Once the object's orientation is computed, one can easily rotate the image as needed.