QR Code Localization Algorithm Based On Convolutional Neural Network

With the recent popularity of devices like mobile phone camera, barcodes have been widely used all over the world as an excellent media for camera to get data. QR code is the most popular and widely used barcode because of its outstanding character storage capacity, fault tolerance and error correction capabilities compared with traditional ID barcode. However, as the application of QR code is becoming increasingly complicated, QR code detection becomes much more difficult so that traditional QR detection algorithm cannot satisfy the requirement anymore. To improve the success rate and speed of QR detection, an efficient and robust QR localization algorithm is needed to extract the correct candidate QR code areas from the input image and conduct distortion correction on them before data decoding.In this paper, we present a QR code localization algorithm based on Convolutional Neural Network (CNN) to detect the candidate QR code areas in complex background efficiently. The main wok of this paper is illustrated as following:Firstly, we utilize sliding window technique to divide the binary image into patches, which are then imported to the convolutional neural network for further classification. These patches will be distinguished and connected to get candidate QR areas. Then, to position QR code more precisely, we propose an algorithm to find the embedded finder patters in the three corners of the QR code quickly. The experimental results show that the QR code localization algorithm in this paper obtains high success rate efficiently. Besides, the proposed algorithm improves precision and stability to a large degree compared with other algorithms based on texture features.