Tag Information Recognition Approaches and Algorithms for Cross-Border Products Checking

The images with fixed layouts, such as images from ID cards, driving licenses, and invoices can be recognized from prior knowledge[1]-[7]. However, The non-immobilized images, such as product labels at ports, is very difficult to be extracted structured data information from tag images because the formats and contents of tags in different countries and different product vary widely[8]. The process is complex and the error rate is high. This paper combines the characteristics of the Cross-Border Products label, overall format complex and simple local structure (top-to-down and left-to-right), and proposes a method for identifying and structuring port commodity label information. The method mainly establishes a template library of keyword and data unit information of commodity labels according to the port commodity classification and then separates the keyword and the data information from the multi-line text with accurate location information recognized by the OCR engine. Finally, the keyword and data are structured according to the local layout pattern between the keyword and the data, and the structured Cross-Border product information is obtained.