Reading Multilevel 2-D Barcodes Using a Machine Learning Approach

Abstract

This paper addresses the reading problem of multi-level 2D barcodes over a Print-and-Capture (PC) channel. The prior reading schemes have different limitations to hinder their applications, e.g., suffering from quantization error, being sensitive to the predetermined decision boundaries, and being sensitive to the selection of initial parameters. In this paper, we introduce a machine learning approach to address the above limitations using a new ensemble clustering algorithm. Based on the new ensemble clustering algorithm, we propose two reading schemes of a multi-level 2D barcode. Specifically, the first proposed scheme is named the Ensemble Clustering (EC) reading scheme. In the EC reading scheme, we introduce a weighted ensemble mechanism to assign different weights to different base clustering results. Then, we propose the second scheme, named the Enhanced Ensemble Clustering (EEC) reading scheme, to further improve the reading performance with the help of the reference symbols. We implement our approach and conduct extensive performance comparisons through an actual excremental platform under various multi-level 2D barcodes and various capturing devices. From experimental results, we observe that both proposed reading schemes have better performance than the prior reading schemes. Moreover, the EEC reading scheme has better performance than the EC reading scheme, and their performance gap becomes more apparent as the distortion of a PC channel increases.