Automated Visual Inspection (AVI): Quantifying Mura and Uniformity

Introduction: Why Mura is Difficult to Define

#

The word “Mura” originates from Japanese, meaning “non-uniformity” or “blemish.” In the display industry, it specifically refers to luminance or chromaticity non-uniformities perceptible to the naked eye on a display panel—those fuzzy patches, cloud-like shadows, or subtle streaks that mar what should be a uniform screen.

The fundamental reason Mura detection is difficult is that it sits at the intersection of physical measurement and human visual perception. A region with a physically minor luminance deviation may not be noticed by the human eye; conversely, the eye is exceptionally sensitive to subtle changes at certain spatial frequencies. Traditional manual inspection relies on inspectors’ experience and subjective judgment, suffering from inherent flaws like poor consistency, low efficiency, and inability to quantify. How to transform this subjective feeling of “it looks wrong” into repeatable, traceable objective data is the core proposition of Mura detection technology.

Classification System of Mura

#

Based on morphological features and location, Mura is typically classified into the following types in the industry:

Classification by Morphology

#

Spot Mura appears as isolated, small-area bright or dark regions. Its causes can include micro-particle contamination, local liquid crystal alignment disorder, or individual pixel anomalies.

Line Mura extends in elongated strips, which can be horizontal, vertical, or tilted. Unlike sharp line defects that traverse the entire screen, line Mura usually has relatively soft boundaries and can be caused by substrate scratches, roller indentations, non-uniform evaporation, and other process factors.

Area/Region Mura refers to luminance or chromaticity non-uniformity regions with larger areas, irregular shapes, and fuzzy boundaries, often described as “cloud spots” or “stains.” This is the most common but also the hardest type to detect because its contrast is usually extremely low and it lacks a fixed pattern.

Clouding specifically refers to large-area, diffused, low-contrast luminance non-uniformity, resembling clouds in the sky. It is usually related to backlight uniformity or wide-range fluctuations in material properties.

Classification by Location

#

Edge Mura appears in the edge areas of the display, possibly caused by light leakage, stress concentration in edge regions, or differences in material properties.

Corner Mura occurs at the corners of the display. For example, “Corner Light” refers to bright spots visible at corners in a dark state, and “Butterfly Mura” is a characteristic dark spot with strong contrast at corners and edges.

Black Mura specifically refers to large-area aggregate patches observed when displaying black or dark screens. This type is particularly critical in the quality evaluation of automotive displays. The German Automotive Work Group Displays has developed standardized measurement methods for Black Mura, published by the German Flat Panel Display Forum (DFF).

Color Mura refers to color deviations on the display rather than luminance deviations. Unlike the detection logic for luminance Mura, color Mura needs to be evaluated in a color space (such as CIE xy or CIE u’v’).

Mura Differences Across Display Technologies

#

The causes of Mura differ fundamentally across display technologies. Mura in LCDs mostly comes from backlight non-uniformity, liquid crystal layer control deviations, and polarizer stress; Mura in OLEDs mainly stems from inconsistencies in pixel self-luminescence and uniformity issues in organic material deposition. MicroLEDs face unique challenges such as mass transfer precision, performance differences in micron-level chips, and splicing gaps. This means detection systems must be adaptable and configurable for different technologies.

Limitations of Manual Visual Inspection

#

Traditional Mura detection relies on experienced inspectors observing screens under controlled lighting conditions, a method with systemic flaws:

Subjectivity and Inconsistency. The human perception threshold for Mura varies individually and is easily affected by fatigue, mood, and ambient light. Standards can be inconsistent between different inspectors or even for the same inspector at different times. Research shows that manual inspection’s coefficient of variation in repeatability and reproducibility is much higher than that of automated solutions.

Efficiency Bottleneck. Manual inspection is time-consuming and labor-intensive, with inspection time for a single panel typically ranging from dozens of seconds to several minutes, making it difficult to meet the Takt Time of modern production lines. The cost of training and maintaining a professional inspection team also continues to rise.

Difficulty in Quantification. Manual inspection results are usually qualitative (pass/fail) or semi-quantitative (slight/severe), failing to provide precise quantitative data usable for process control and quality traceability. For example, an inspector can judge that there is “Mura” somewhere but cannot give its precise contrast value or spatial frequency characteristics.

Risk of Missed Detection. For area Mura with extremely low contrast, even experienced inspectors may miss them after high-intensity work. Yet, such low-contrast Mura is often one of the quality issues most sensitive to end-users.

JND Model: The Bridge Connecting Physical Measurement and Human Perception

#

What is JND?

#

The Just Noticeable Difference (JND) is a fundamental concept in psychophysics, referring to the minimum amount a stimulus (such as luminance or color) needs to change to be noticed by approximately 50% of observers. In Mura detection, JND = 1 is defined as the threshold just perceivable by a typical observer.

The introduction of JND solves a fundamental problem: whether a physical deviation measured by an automated system truly constitutes a “visible defect.” A region with a physical luminance difference of 0.5 cd/m² might be very conspicuous against a low-luminance background (JND > 1) but might be completely imperceptible against a high-luminance background (JND < 1).

How the JND Model Works

#

JND models are built based on key characteristics of the Human Visual System (HVS):

Contrast Sensitivity Function (CSF). The human eye has different sensitivities to luminance changes at different spatial frequencies. Sensitivity is highest in the range of approximately 2~5 cycles/degree; it drops significantly at very high or very low frequencies. The CSF can be characterized as a band-pass filter whose center frequency and bandwidth change with background luminance.

Standard Spatial Observer (SSO) Model. Developed by agencies like NASA, the SSO model comprehensively considers CSF, aperture effects (the eye being more sensitive to the center of the visual field than the edges), and orientation selectivity. it can predict the perceptibility of Mura based on its physical characteristics (size, contrast, spatial frequency, orientation) and viewing conditions (background luminance, viewing distance).

Application of JND in Mura Quantification

#

In practice, a typical workflow for JND analysis is:

1. Use an imaging colorimeter to accurately capture the spatial luminance distribution map of the display.
2. Pre-process the luminance image, including noise removal, background correction, and Moire removal.
3. Input the processed luminance data into a JND analysis algorithm, which calculates JND values for each pixel position based on the HVS model.
4. Generate a JND Map, intuitively showing the distribution of perceivable Mura across the entire screen.
5. Perform automated pass/fail judgment based on preset JND thresholds (e.g., JND > 1 is considered a failure).

The core value of JND analysis is transforming the judgment of “whether there is a defect” from relying on subjective human feeling to objective calculation based on visual science models. Using the same JND threshold standards, results across different production lines and time periods are directly comparable.

Workflow of Automated Visual Inspection (AVI)

#

A complete automated visual inspection system for Mura typically includes the following core steps:

Image Acquisition

#

An Imaging Luminance Measurement Device (ILMD) is used as the core acquisition equipment. An ILMD is essentially a high-performance industrial camera with precise spectral calibration, equipped with CIE-standard filters, capable of simultaneously capturing the luminance and/or chromaticity distribution over the entire display surface.

Key technical requirements for the imaging equipment include:

• Resolution: Camera pixel count should be no less than the display pixel count (reproduction ratio >= 1), ensuring each display pixel corresponds to at least one camera pixel.
• Spectral Match: The camera’s spectral response must match the human spectral luminous efficiency function V(λ), with a characteristic value f1’ less than 5%.
• Flat-Field Uniformity: The camera’s own non-uniformity f21 should be less than 2%; otherwise, the camera’s non-uniformity will be misjudged as Mura.
• Dynamic Range: Sufficiently high dynamic range is needed to capture weak Mura signals in low-luminance frames.
• Temporal Stability: Measurement value differences should be less than 0.1%.

Geometric Alignment and Calibration

#

Strict device alignment is required before measurement: the display should be perpendicular to the camera lens, centered, with rotation and tilt errors controlled within 0.5 degrees. Specialized test patterns (such as CROSS or Grill 4_4 patterns) assist in alignment and focusing.

Regarding Moire interference—aliasing between the camera sensor array and the display pixel array—suppression is achieved through moderate defocusing or specialized filtering algorithms. The Black Mura standard recommends defocusing until the modulation depth of the grill test pattern reaches 50%~90%.

Image Processing and Analysis

#

The captured raw images undergo multiple levels of processing:

Pre-processing stage includes flat-field correction (eliminating camera non-uniformity), black-level correction (suppressing dark current noise), and linearity correction (ensuring pixel values correspond linearly to actual luminance).

Mura feature extraction can take several technical paths. Traditional methods include spatial domain filtering, frequency domain analysis (Fourier transform, wavelet transform), Gabor filters, etc. In recent years, deep learning-based methods (such as CNN, U-Net semantic segmentation) have shown superior performance in handling low-contrast and irregularly shaped Mura.

Quantitative evaluation typically includes:

• Virtual Spot Area Scan Method: Simulates a standard luminance meter by moving a rectangular measurement frame across the display area with single-pixel increments, calculating local average luminance and evaluating overall uniformity.
• Luminance Gradient Calculation: Calculates the first derivative of luminance (absolute and relative gradients) to quantify the spatial luminance change rate.
• JND Analysis: Calculates perceptibility based on human visual models.

Judgment and Output

#

The system automatically makes OK/NG judgments based on preset threshold standards. For example, in the Black Mura standard, the luminance difference between the center and the four corners should not exceed 5%, and the maximum relative luminance gradient should not exceed a specific threshold. Evaluation results are output as tables and pseudo-color image maps, saved as standardized CSV or image files for subsequent statistical analysis and quality traceability.

Uniformity Evaluation Metric System

#

Quantification of Mura detection relies on a complete system of uniformity evaluation metrics:

Luminance Uniformity

#

Multi-point Sampling Method. The most basic method involves selecting several sampling points on the screen (e.g., 5-point or 9-point methods in the VESA FPDM standard) to measure luminance, measuring uniformity by the ratio of maximum to minimum values or by standard deviation. This method is simple and fast but has low spatial resolution and easily misses local Mura.

Full-screen Scanning Method. Uses imaging equipment to analyze the entire screen pixel by pixel, capable of generating a complete uniformity distribution map. The virtual spot area scan method belongs to this category; its measurement frame size (e.g., 10x10 or 11x11 pixels) can be adjusted as needed, directly affecting spatial sensitivity of detection.

Luminance Gradient

#

Luminance gradient reflects the spatial rate of change of luminance and is a key metric for characterizing Mura boundaries:

• Absolute Luminance Gradient: The absolute value of the first spatial derivative of luminance, in units of cd/m²/pixel.
• Relative Luminance Gradient: Calculated by dividing the absolute luminance gradient by a reference luminance (such as the average luminance of a white or black field) to obtain a dimensionless ratio.

Relative gradient is particularly important when evaluating Mura under dimming conditions—the same absolute luminance difference is much more conspicuous under low backlight conditions than under high backlight conditions.

Chromaticity Uniformity

#

In the CIE 1976 u’v’ uniform chromaticity space, Delta u’v’ (the length of the chromaticity deviation vector) is used to measure chromaticity uniformity. Delta u’v’ = 0.002 is typically regarded as the level of the perception threshold for chromaticity differences. For strict applications (e.g., medical displays), thresholds may be set within 0.004.

SEMI Mura Quantification Formula

#

The SEMI standard proposes specific formulas for Mura quantification. The grayscale difference ratio Cx is calculated based on the difference between the average grayscale of the Mura area (Im) and the background area (Ib); the grayscale contrast value Cjnd based on JND further introduces human visual perception factors. These quantitative metrics provide a unified “language” for Mura assessment across different devices and production lines.

Black Mura Standard: Strict Requirements for Automotive Displays

#

The Black Mura standard is a specialized test method developed by the German Automotive Work Group Displays, targeting uniformity issues of automotive displays in dark (black field) conditions. Its core design philosophy is to relate technical measurement with human perception.

The test workflow of this standard is highly structured, managed through “Projects” and “Recipes,” where recipes contain detailed thresholds and settings for various evaluations. The standard requires measurement using specific test patterns like pure white, pure black, black with white borders, stripes, and Grill 4_4, covering multiple dimensions such as center-corner comparison, resolution check, focus evaluation, and uniformity and gradient assessment.

The Black Mura standard holds a special position in the automotive industry because automotive displays often need to work under low-luminance conditions (e.g., night driving), where uniformity defects in dark states are particularly disruptive to the driver’s visual experience. The evaluation methods specified in the standard comprehensively consider CSF and JND, ensuring measurement results are highly consistent with the driver’s actual perceptual experience.

From Inspection to Process Feedback: The Deeper Value of Mura Data

#

The value of Mura detection lies not only in end-of-line quality screening but also in providing feedback signals for manufacturing processes. By accurately classifying and statistically analyzing detected Mura, engineers can trace the root causes:

• Regularly occurring spot Mura at specific locations might point to equipment contamination sources.
• Line Mura distributed in specific directions might reveal roller pressure anomalies.
• Large-area clouding Mura might reflect characteristic fluctuations between material batches.
• Edge or corner Mura might be related to stress control in the assembly process.

This root cause analysis based on Mura classification transforms the inspection system from a passive “gatekeeper” into an active process improvement tool. In the era of OLED and MicroLED, this idea further extends to Demura compensation, where the inspection system no longer just judges pass/fail but participates directly in the closed loop of improving product yield—which is exactly the subject that the next article will explore in depth.

FAQ

#

Q1: What role does the JND model play in Mura detection?

#

The JND (Just Noticeable Difference) model bridges physical measurement and human visual perception. Based on characteristics of the Human Visual System such as the Contrast Sensitivity Function (CSF), it calculates whether luminance deviations at each position on a display panel are perceptible to the human eye. JND=1 represents the threshold just perceivable by approximately 50% of observers. Through JND analysis, automated inspection systems can determine whether a physically existing luminance deviation constitutes a ‘visible defect,’ transforming subjective visual judgment into objective calculation based on visual science models, making results directly comparable across different production lines and time periods.

Q2: What are the key technical requirements for imaging equipment in an AVI system?

#

An AVI system imposes five key requirements on Imaging Luminance Measurement Devices (ILMDs): resolution must be no less than the display’s pixel count (reproduction ratio >= 1); spectral match characteristic f1’ must be below 5% to match the human spectral luminous efficiency function V(lambda); flat-field uniformity f21 must be below 2% to prevent the camera’s own non-uniformity from being misjudged as Mura; dynamic range must be high enough to capture weak Mura signals in low-luminance frames; and temporal stability must keep measurement differences below 0.1%.

Q3: Why is the Black Mura standard particularly important for automotive displays?

#

The Black Mura standard was developed by the German Automotive Work Group Displays specifically for uniformity issues in automotive displays under dark (black field) conditions. Its special importance stems from the fact that automotive displays frequently operate under low-luminance conditions such as night driving, where uniformity defects in dark states are particularly disruptive to the driver’s vision. The standard comprehensively considers the human Contrast Sensitivity Function (CSF) and JND models, linking technical measurement with human perception to ensure results are highly consistent with the driver’s actual perceptual experience, covering dimensions like center-corner comparison, uniformity, and gradient assessment.

This article is part of the Imaging Colorimeter Technology Knowledge Base series.