Can You See What Others See –A Defect Detection Model for Patterned Backgrounds

Abstract

No one wants defects on their products – neither manufacturers nor consumers. For manufacturers, a series of quality inspection processes is always required to ensure qualities of all their products are under control. For consumers of liquid crystal displays (LCDs), however, they may only concern whether they can see defects on their displays. In this sense, visibility of defects is more important than whether they really exist. Mura is such type of detects on LCDs that most people might have neither heard about it nor been aware of (existence of) them on their panels but some professional users do concern them a lot. This reflects that Mura defects are visually hard to notice and thus tend to be easily overlooked by most users; but in certain circumstances they are really an issue to image quality. In this research, a Mura detection model based on human visual perception was established particularly for visibility prediction of Mura on patterned backgrounds. It is also a model different from current industrial standards which suggest, and therefore are limited to, inspections of Mura defects to be carried out on uniform neutral grey backgrounds. Our analysis showed that colour did not have as much influence on the visibility as previous studies reported when Mura defects are viewed against patterned backgrounds. For a given Mura size and special frequency of a patterned background, there existed a linear relationship between the model outputs (dR, in terms of ΔE*ab) and the lightness (L*) of the background. An interesting phenomenon was also found that the 1st derivative of the slopes (i.e. slope variation across different experimental conditions) of the linear functions representing the relationships mentioned above can represent the special frequency effect whilst the 2nd derivative of the slopes represents the Mura size effect on the visibility of Mura patterns. Apparently this model provides more reliable predictions of visibility to situations closer to the reality that it is usually complex images, rather than uniform colour patches, are displayed on LCDs. Preliminary analysis also shows that the proposed model can deliver more reliable results for patterned backgrounds than S-CIELAB does.