Dynamic OCR Inspection

Production processes in the pharmaceutical industry are subject to particularly strict legal requirements. According to the guidelines of Good Manufacturing Practice (GMP), products must be manufactured consistently to high quality standards. Accordingly, the specifications place high demands on the documentation of processes and the testing of goods. Regular inspections are intended to ensure maximum safety and the effectiveness of the products.

The global health and food group H&H Group is also directly affected by GMP: In order to reliably comply with the regulations, the manufacturer must correctly label the packaging of its medical and cosmetic products. For this purpose, various information such as batch number, production date and expiration date are printed on the cartons. The combinations of characters must be correct and complete. This is the only way to correctly assign the products and consistently meet the quality requirements with regard to labeling. It is therefore essential to check the character codes carefully before the cartons leave the factory and are delivered to the customer.

Insufficient controls led to complaints

Previously, this inspection process was only carried out manually at the H&H Group, which proved to be very costly and time-consuming. Due to the high number of packs and the fast cycle time, the inspection could often only be carried out on a random basis. In some cases, this led to incorrectly labeled products being delivered to wholesalers and other customers. As a result, complaints and objections became more frequent, leading to financial losses, customer losses and serious damage to the company's image.

In order to improve the quality of product labeling, the inspection process had to be made more efficient. The managers at H&H Group Shenzhen therefore commissioned Jiangrun Xuneng Technology to develop a suitable solution. The company specializes in 2D and AI software frameworks and focuses primarily on 2D and 3D vision solutions, including localization, identification and measurement of objects. The provider serves a wide range of industries, such as food, medicine, chemicals, packaging, automotive and rail transportation.

Machine vision platform automates visual inspection

Shenzhen Jiangrun Xuneng Technology has developed a machine vision platform that fully automates the visual inspection of product markings on packaging. The combinations of letters and numbers are read and checked using modern OCR (Optical Character Recognition) technologies. "This application presented us with major challenges. The characters printed on the cartons are usually encrypted, with different code types and sizes being combined. In addition, the characters are often not exactly in the intended position, but are offset. In addition, letters and numbers are often printed incorrectly, unclearly or blurred on the packaging, which makes it difficult to clearly identify the codes," explains King Wang, Technical Application Engineer at Jiangrun.

Another challenge is the high cycle time in the test sequence. The characters must be read and checked at high speed in order to achieve maximum throughput. This requires comprehensive pre-processing by the OCR algorithms. This is the only way to reliably find the respective target area and ensure the highest possible recognition accuracy, even with different character types and sizes.

Precise identification and localization of characters

With Jiangrun's machine vision platform, these complex requirements can be precisely implemented. The standard machine vision software product Halcon from MVTec, which forms the core of the solution, is used here. The Halcon algorithms are used for the pre- and post-processing of images as well as the precise identification and localization of characters.

First, the software must precisely recognize the character background in the image and find a horizontal reference line to determine the exact position of the characters. This involves defining a region of interest (ROI) within which the numbers and letters, such as an expiry date or a batch designation, are recognized, read and checked. Errors in the marking can be reliably detected by comparing them with reference characters.

"Thanks to MVTec Halcon, we were able to significantly shorten the project duration, reduce the costs for development, operation and maintenance of our machine vision platform and noticeably increase implementation efficiency. We can also implement dynamic OCR with high stability and processing speed. New product types, markings or fonts can be flexibly integrated into the application. In this way, we can optimize OCR inspection, cover a wide range of different character specifications and also achieve good results in the event of display errors or incorrectly positioned characters," says King Wang.

The end customer also benefits from the machine vision platform implemented with MVTec Halcon: the H&H Group can automate the entire inspection process and relieve employees of monotonous manual work. This increases the efficiency of the inspection processes and keeps the quality of the product markings at a consistently high level. This reduces the number of complaints to a minimum and significantly strengthens the brand image.

This article is based on documents from MVTec.