Z. Hamici
Pattern recognition, super-sampling, circular cross-correlation, geo-metric transformations
This paper presents a novel algorithm for two-dimensional pattern recognition of binary images. The recognition is invariant to rotation, translation, and scaling. The algorithm permits estimation of the geometric transformations with high accuracy. The 2-D pattern recognition problem is converted to one-dimensional signal recognition using circular cross-correlation. The contour signal is generated from each pattern using Freeman edge tracking, and a one-dimensional signal is generated from the centre of mass to the edge for each object. The circular cross-correlation requires signals with the same length, whereas patterns with different scale have a different number of pixels in their contours; hence the super-sampling is introduced to increase the number of samples in the lower-resolution contour signal. The algorithm, tested on simulated binary images, shows excellent results in recognizing rigid patterns with complex or arbitrary shapes. The low computational requirement of the algorithm permits this technique to be used by manufacturing robotics and flexible automation systems that require real-time product management. The recognition process and geometric transformation parameters estimation performed on images with 256 × 256 resolution are obtained in less than one second.
Important Links:
Go Back
