A Hybrid Character Recognition Approach Using Fuzzy Logic and Stroke Bayesian Program Learning With Naïve Bayes in Industrial Environments

Automated character recognition is critical for reading and tracking data in a variety of fields. It is particularly challenging in industrial settings since information may be printed on the surface of various materials with complex and uneven shapes, causing overlapping, obstructing, and distorting characters. We propose a hybrid character recognition approach using fuzzy logic and stroke Bayesian program learning with naïve Bayes. During character segmentation, touching characters are separated using support vector machines and a three-feature fuzzy segmentation strategy that uses particle swarm optimization. This approach includes a new methodology for stroke presentation and extraction using a prebuilt primitive-stroke library containing prior knowledge. During character recognition, a conceptual character model is constructed using stroke Bayesian program learning. Monte Carlo Markov chain sampling is used to produce a fitting model for each character. This model predicts character classification by calculating the probability that a target sample belongs to a training set. To this end, naïve Bayes effectively discerns extremely similar characters. We evaluate our method experimentally using a database of industrial images and the NIST dataset. Our method outperforms existing state-of-the art methods.