Automatic Detection of Reference Region in Label Images
Yuheng Wang, Chune Zhang · 2016
In regard to automatic quality inspection for label printings, a reference region is necessary for finding the region of interest.With such a reference region, the defects of the region of interest can be analysed.The reference region should be unique, stable and have a high contrast against the background.This paper proposes a method to automatically locate such a reference region.First, the maximally stable extremal regions method is used to obtain a series of regions, including reference regions and noisy regions.Next, a constrained model is established to filter out regions that are undesirable for use as reference regions.After that, to ensure all filtered regions are unique, a method to detect similar regions is applied based on shape-matching.Compared with other similarity measure methods, the shape-matching method shows better performance.Experiments on label images show the effectiveness and stability of our algorithm.The proposed method provides a new way for quality inspection of label printings in an industrial setting.