Bottle Printing Defect Detection Based on CUDA Accelerated Hybrid Image Registration

Objective: in order to detect the printing defects of the legal text of aluminum cans and beer bottles, the difference method after image registration based on difference model is used. Aiming at the problems of insufficient accuracy and slow speed of text image registration, a method of accelerating mixed image registration based on CUDA is proposed. Methods: firstly, rough registration based on speeded-up robust features was used to make the detected image basically coincide with the template image; Secondly, the local deformation fine-tuning based on demons elastic registration algorithm is used to improve the registration accuracy; In order to improve the speed of registration, NVIDIA CUDA architecture is used to accelerate and optimize the algorithm in parallel to complete the registration quickly. The improved hybrid registration method can quickly and accurately complete the registration with the template image. After the differential processing, the artifacts are eliminated by threshold segmentation and morphological corrosion expansion processing to complete the defect detection. Results: the experimental results show that the registration accuracy of this method has been significantly improved, and the CUDA acceleration can achieve about 24 times of the acceleration ratio, which greatly shortens the registration time and meets certain detection requirements. Through the detection of 63 defect samples, the success rate is 98.41%. Conclusion: the hybrid registration scheme proposed in this paper can accurately and efficiently complete the bottle label text registration, and realize the defect detection task, which provides an effective solution for the printing defect detection of the legal text of aluminum cans and beer bottles.