Segmentation-free printed character recognition by relaxed nearest neighbor learning of windowed operator

Segmentation is considered by many researchers as the key technology for a reliable optical character recognition (OCR) system. To accomplish sound segmentation, many alternative techniques have been recently proposed. This paper presents a new technique to recognize characters without explicit segmentation. It is based on the automatic construction of a windowed operator by relaxed nearest neighbor learning. It has been implemented, tested and yielded excellent recognition accuracy and computational performance.