Research On Automatic Detection For Reference Region In Surface Quality Inspection

With the rapid development of the printing industry,printing companies are paying more attention to product cost and product quality control.Automatic inspection has become the main way of inspection in printing industry.Before an inspection work,workers usually set a qualified image of sample as a template.The sample image will be captured in real-time and then compared with the template images for registration and analysis.In order to ensure the accuracy of registration,in the template image,workers usually select a significant and unique area to assist in registration.This region is called a reference region.Traditional method to select a reference region is manual selection.This method will bring a lot of problems.First of all,subject to a priori knowledge,reference regions selected by different workers are not the same.There is no clear definition for reference region.The select quality varies among different workers;Training workers for reference region selection will virtually increase the cost of automated detection system.In this paper,based on the electronic product label,we focus on the automatic detection for reference region and present some feasible methods as follows:(1)In this paper,we collect electronic products in daily lives,simulate the industrial scene environment to capture the electronic product label images,set up a label image database to provide the sample for the following experimental analysis.(2)In this paper,the definition and selection criteria of a reference region are given.At the same time,a method is proposed to extract candidate reference region of label image.Firstly,the connected component is partitioned based on the maximum stable extreme region algorithm.Secondly,a constrained model is established with four parts,such as area,variance,aspect ratio,and boundary distance,and the regions which are not suitable to be reference regions are filtered out.Then the adjacent region merging criterion is designed and the neighboring regions are merged to ensure the integrity of the region.Finally,the product of the gradient of different regions is used in the similarity measure,the similar region is marked and deleted.Experiments show that the algorithm is simple and effective,in addition,it is not sensitive to tiny mechanical system deviation.(3)In this paper,based on the study of the characteristics of the reference region,an optimal reference region extraction method is designed.First,the region features that represent the reference region are designed and extracted.Next,the optimal reference region criterion is designed based on the outlier detection.The optimal reference region is obtained by the two region features.Finally,based on the obtained optimal reference region,a series of experiment are designed to test the stability and registration time of reference region.Experiments show that the optimal reference region can replace the artificial selection.The time consuming is low and the reliability is high.