A computer vision approach to digit recognition on pulp bales
This paper describes a computer vision approach for recognizing quality and producer information of pulp bales from digit series stamped on pulp bales. The digit recognition consists of three stages: segmentation of digit series, feature extraction, and classification. Segmentation of digit series is based on image thresholding and Randomized Hough Transform. Digit segmentation produces six digit windows. In feature extraction two band-pass derivative of Gaussian filters are used and the resulting gradient field histograms are used after normalization in classification of digits. The digits in the test set can be classified 93% correct with a multiple layer perceptron network. Classification results with three other well known classifiers are also reported.
Bibliographic Reference: Article: Pattern Recognition Letters, Vol. 17 (1996), pp. 413-419
Record Number: 199610868 / Last updated on: 1996-09-16
Original language: en
Available languages: en