Linear Discriminant Analysis Cauchy Estimator for Single Chinese Character Font Recognition

Abstract

In machine learning area, the rapid development of optical character recognition (OCR) has prompted interest in Chinese character font recognition (CCFR), especially single Chinese character font recognition. However, when pre-processing pollutes Chinese font images with noise, traditional font recognition algorithms tend not to be suitably discriminant. In this paper, and based on linear discriminant analysis and Cauchy estimator theory, we propose a novel feature selection algorithm called linear discriminant analysis Cauchy estimator (LDACE) for single Chinese character font recognition. LDACE aims to: (1) consider both between-class and within-class geometry in the low-dimensional space, and (2) preserve recognition when input samples are polluted with noise. Experiments with the frequently used FUCCFR dataset demonstrate LDACE’s effectiveness.