A Multiclassification Method for Iris Data Based on the Hadamard Error Correction Output Code and a Convolutional Network

Abstract

Although there have been many successful multiclassification models in the field of image recognition in recent years, adding new classes to the trained models remains a research focus. In this paper, we present a multiclassification method based on the Hadamard error correction output code and a convolutional neural network to solve this problem. Compared with other multiclassification methods, it not only makes use of the excellent image processing performance of a convolutional network but also combines the Hadamard error correction code's characteristics of simple construction and adaptability to arbitrary categories. The characteristics of adding new classes are shown in the experiment. We applied this model to the iris multiclassification problem and compared it with the traditional classifier. The experimental results show that our method achieves accuracies of 98.19% and 96.35% on the CASIA-Iris-LampV4 and the JLU-4.0 iris datasets, respectively, which is better than other classic classification models, and it can effectively add new categories with only partial parameter modification.