摘要

Visual defects in liquid crystal display images often appear as low contrast and blurry contour without distinct intensity difference from their surrounding region. Besides, the background is usually intensity inhomogeneity. All these properties make the machine vision inspection extremely hard. This paper presents an effective machine vision inspection method using a local active contour model to detect defects with different brightness levels as well as diverse sizes and shapes. A modified local binary fitting model which is robust to initial contour is developed to extract defect boundary. Meanwhile, a simple preprocessing scheme is given to compensate the drawback of the two-phase active contour model for detecting objects with wide brightness levels. Experimental results show that the presented method can detect various types of defects effectively and achieves high performance in terms of inspection accuracy (both precision and recall are higher than 0.99).