This paper presents a vision-based recognition system for rotated characters using artificial neural networks (ANN), trained by back-propagation algorithm. The geometric features of the captured image of the input character are generated using the rapid transform (RT) algorithm. The RT produces the features of the input character that are invariant to rotational transformation. The input character image is represented by a sequence of sectors defining its components. The sequence that corresponds to a rotated character input can be determined from an original sequence of the same input character through a series of cyclic shifts. The resulting features are then fed to the ANN for learning and recognition of characters. To determine the orientation of an input character, its corresponding features are fed to a specific ANN to determine the value in degrees. This system presents a character recognition and orientation determination solution in real-time. The results show that it is able to process and recognize the images at an acceptable speed of 12 fps. The experimental results show that the algorithm for character recognition recognized 91% of the characters and the orientation determination correctly determined the approximate orientation of 74-81% of the characters.
展开▼
