White Spot Mura Defect Detection in Display Panels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated detection of Mura defects in displays, particularly those with low contrast and no clear edges, is challenging due to the difficulty in distinguishing defects from their surroundings and uneven illumination, making manual detection time-consuming and ineffective.
Innovation Solution
A system and method that preprocesses images by generating image patches centered on local maxima candidates, extracts feature vectors including image moments and texture features, and uses a machine learning classifier, such as a support vector machine, to classify defects like white spot Mura, with noise filtering and Gaussian smoothing to enhance detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated surface inspection is used to detect defects, then detection speed and productivity are improved, but detection accuracy deteriorates for low-contrast defects like Mura that have no clear edges and uneven illumination
Solution Approach 1:
The inspection system divides the display screen into multiple overlapping image patches and processes each patch independently through local maxima detection and classification. This segmentation allows the system to handle large-area low-contrast Mura defects by focusing on local brightness variations in manageable units, thereby maintaining both high processing speed and accurate detection of subtle defects.
Solution Approach 2:
The system applies local contrast enhancement and texture analysis specifically to regions containing local maxima candidates, rather than processing the entire image uniformly. By focusing computational resources on suspicious local regions with enhanced contrast and texture features, the system achieves high detection accuracy for low-contrast Mura defects while maintaining overall processing efficiency.
2Measurement precision
If manual defect detection is used to achieve high accuracy for low-contrast Mura defects, then measurement precision is improved, but productivity deteriorates due to time-consuming inspection
Solution Approach 1:
The system performs preliminary processing by generating image patches and identifying local maxima candidates before applying complex classification. This preliminary action filters out obvious non-defect regions, allowing the detailed texture and moment analysis to focus only on suspicious areas, thereby achieving manual-level accuracy at automated speeds.
Solution Approach 2:
The system replaces manual visual inspection with an automated computational approach that uses image processing algorithms, local maxima detection, and machine learning classification. This substitution eliminates the time-consuming nature of manual inspection while maintaining or exceeding human detection accuracy for Mura defects.
3Measurement precision
If the entire input image is processed uniformly to detect defects, then measurement precision is improved, but device complexity and processing time increase
Solution Approach 1:
The system segments the input image into multiple patches and processes each patch independently through a standardized pipeline of local maxima detection, feature extraction, and classification. This modular segmentation reduces overall system complexity by breaking down the complex task of full-image defect detection into simpler, repeatable unit operations.
Solution Approach 2:
The system applies full processing (local maxima detection, feature extraction, classification) only to image patches that contain local maxima candidates, rather than uniformly processing every patch. This partial action approach reduces computational complexity and processing time while maintaining detection accuracy by focusing resources on suspicious regions.
Data Source
Figure 1A
Figure 1B
Figure 2
AI summary
A system and method for white spot Mura defects on a display. The system is configured to pre-process an input images to generate a plurality of image patches. A feature vector is then extracted for each of the plurality of image patches. The feature vector includes at least one image moment feature and at least one texture feature. A machine learning classifier then determines the presence of a defect in each patch using the feature vector.