An Algorithm for the Recognition of Motion-Blurred QR Codes Based on Generative Adversarial Networks and Attention Mechanisms

Motion blur can easily affect the quality of QR code image, making it difficult to recognize QR codes on moving objects. This paper proposes an algorithm for the recognition of motion-blurred QR codes based on generative adversarial network and attention mechanism. Firstly, a multi-scale feature extraction framework for motion defuzzification is designed using deep convolutional neural networks, and enhanced multi-scale residual blocks and multi-scale feature extraction modules are utilized to capture rich local and global features. Secondly, the efficient channel attention module is added to strengthen the weights of effective features and suppress invalid features by modeling the correlations between channels. In addition, training stability is achieved through the use of the WGAN-div loss function, leading to the generation of higher quality samples. Finally, the proposed algorithm is evaluated through qualitative and quantitative comparisons with several recent methods on both the GOPRO public dataset and a self-constructed QR code dataset, respectively. The experimental results demonstrate that, compared to the other methods, the proposed algorithm has shown significant improvements in both processing time and recognition accuracy when dealing with the task of severe motion-blurred QR code recognition.