A hybrid approach based on Hotelling statistics for automated visual inspection of display blemishes in LCD panels

This paper presents an alternative hybrid approach based on Hotelling statistics, combining ant colony method and neural network model to automatically detect the display blemishes in liquid crystal display (LCD) panels. Owing to their space saving, energy efficiency, and low radiation, LCD’s have been widely applied in many high-tech industries. However, the display blemishes such as abnormal spots (white and black spots) and slight color variations (bright and dark regions) often exist in LCD’s. To detect these color unevenness blemish detection, this research proposes a multivariate statistic based hybrid defect detection approach. We first use multivariate Hotelling statistics to integrate different coordinates of color models and construct a Hotelling distance diagram to represent the degree of color variations for selecting suspected blemish regions. Then, an ant colony algorithm that integrates computer vision techniques precisely identifies the abnormal spot defects in the Hotelling distance diagram. And, the back propagation neural network model determines the regions of slight color variation blemishes based on the Hotelling distance values. Experimental results demonstrate the validness and efficiency of the proposed approach.