Industrial character recognition based on grid feature and wavelet moment

Abstract

A novel descriptor for laser printing character recognition is proposed by using the fusion features and multilevel classification. There are two level features in the feature extraction module, the first level feature uses the coarse grid statistical features, having great applicability for local character distortion and stroke thickness inequality. Wavelet moment, as the second level feature, has the scale, translation and rotation invariance and the great anti-noise capability. It can reflect the approximation of the overall character and local detail information at the same time, so it is suitable for classing similar characters. As for the classification module, based on results of primary feature template matching classification in the first place, the second matching uses two level features fusion and special distortion character template to classify the top three character in the results. Experimental results demonstrate that using the fusion features and multilevel classification, industrial laser character recognition rate can be up to 99.2%, which is better than that of using single stage feature recognition.