Determining object orientation from a single image using multiple information sources

Three techniques are presented which use assumptions about the real world to determine the orientation of objects from a single visual image. The orientation information from each of these techniques is combined to provide a more accurate estimate of object orientation. This algorithm is applied to the task of estimating the orientation of a single transistor against a uniform, but contrasting, background. Three techniques are proposed for estimating object orientation. Histogram Template Matching employs a nearest-neighbor classifier using the normalized correlation function as a distance measure between the histogram of the input image and a set of training histograms. Binary Connectivity Analysis analyzes the connectivity of an object's silhouette and uses the resulting image features to determine orientation. Ellipse Fitting uses the parameters of an ellipse in the image to specify the orientation of the corresponding circular object surface. Location of the image ellipse is accomplished by exploiting knowledge about object boundaries and image intensity gradients. The orientation information from each of these three methods is combined using a "plausibility" function. This probability-based, sub-optimal, decision rule employs weighted sums of joint conditional probabilities to enhance robustness. The combined techniques for estimating orientation were tested on 138 images of transistors. The limitations of this research, and suggestions for further study, are also discussed.