Research On PDF417Barcode Detection And Recognition Methods

In recent years, the surge of intemet-of-things and mobile internet brings new opportunities for the development of barcode. The2D barcodes, such as QR code and PDF417code, have been widely used in our daily lives. Therefore, detection and recognition the2D barcodes have been the hotspot among the industry and academic fields, drawing quite a lot of attentions. This thesis is dedicating to propose several new methods on PDF417barcode detection and recognition. These methods are proposed based on extensive analysis of the up-to-date algorithms. The main contributions of this thesis are listed as following.First, a new PDF417barcode detection method based on Gabor texture description and the SVM classifier is proposed. Current detection methods mainly use the morphological operations to locate the rectangle areas in the image. However, when dealing with images with corrupted PDF417barcodes, these methods fail to work. To solve this problem, we utilize the texture characteristics to locate the barcodes. In detail, multi-scale and multi-direction Gabor filters are used to extract the texture features of each pixel. Then these texture features are used by K-means cluster algorithm to segment the image into several barcode candidate regions. SVM classifier is trained to tell whether a candidate region is PDF barcode or not. The experimental results proved that the proposed method is effective.Second, a sub-pixel based method is proposed to help get better binary result of PDF417barcode region. This method is devised specially for the low-resolution PDF417barcode images. In such a scenario, it is often quite difficult to obtain the correct barcode word, and the proposed sub-pixel interpolation will enrich the black/white distribution information of the barcode area. Experimental results also show that the proposed method could improve the decoding accuracy.Third, an integrated PDF417barcode system, composed by an encoding module and a decoding module, is designed with modularize structure and implemented using C++/MFC/OpenCV. This system could decode PDF417images with high accuracy and in real time. The system is apt to be revised and more methods could be added in it, thus it would play an important role in our future research.