Ensemble classifier and its application to image-based MICR character recognition

Image-based magnetic ink character recognition (MICR) is a challenging research topic in the automatic check processing. In this paper, a novel ensemble classifier system, which consists of three artificial neural networks (ANNs) and a gating network, is used to congregate the recognition results in order to increase the recognition rate and reliability at the same time. A fast and efficient scheme of the genetic algorithm used to evolve the weights of the gating network is presented. A new bending line detection algorithm for the check image processing is proposed. The position information of the detected lines is utilized to connect the broken lines caused by the bending line problem and to enhance segmentation accuracy. The experiments demonstrated that the proposed ensemble classifier system not only increased the overall recognition performance, but also introduced a rejection strategy to suppress the misrecognition rate.