The Hough transform is used to form feature vectors corresponding to the existence of line segments of certain orientations within each image. Results show that these features are a useful addition to other more common OCR features. The author explores the interpretation of all stages (edge orientation detection and subsequent HT) as neural network type operations. The advantage is that all stages of the process become trainable, and one can develop Hough transforms that become expert at spotting the features most common in each class of character.
展开▼
