The authors present a method of measuring the degree of similarity of handwritten characters. This method is based on the segments of handwritten characters. Each handwritten image is decomposed into several segments. Then they compare the segments which belong to different handwritten images using their distance fields. The result is distance of two handwritten images. These distances can be used in training a neural network for the task of handwritten character recognition. In order to make clear segments from handwritten characters, before the measurement, a hardware oriented thinning algorithm is introduced. Every pixel in a handwritten image can be processed in parallel to save time.
展开▼
