Multi-agent-based Two-dimensional Barcode Decoding Robust against Non-uniform Geometric Distortion

Abstract

Two-dimensional (2D) codes are subject to distortion when printed on non-rigid materials, such as papers and clothes. Although general 2D code decoders correct uniform distortion such as perspective distortion, it is difficult to correct non-uniform and irregular distortion of the 2D code itself. To overcome this problem, an agent-based approach is presented here to reconstruct the 2D code. In this approach, auxiliary lines are placed on a 2D code and used to recognize distortion. First, 2D code area is identified through feature patterns composed by the auxiliary lines, and Convolutional Neural Network (CNN) is used to discriminate the patterns. Then, many agents simultaneously trace the lines referring to the various image features and the neighborhood agents. The feature weights are optimized by Genetic Algorithm. The experimental results indicate that agents successfully tracked auxiliary lines right up to occluded area, and the proposed method could decode distorted 2D codes. The performance of the proposed method against distortion level and occlusion amount was also clarified.