The Algorithmic Eye: Leveraging Blob Analysis and Region of Interest (ROI) Tools

The value of an imaging colorimeter lies not only in its precision optical hardware but also in the ability of its companion image processing software to transform raw photometric data into actionable quality decisions. This article will introduce the core functional architecture of imaging colorimeter software, focusing on key features such as Region of Interest (ROI) definition, Point of Interest (POI) auto-detection, Blob analysis, false color mapping, and data export in the context of display defect identification.

I. Core Functional Architecture of Imaging Colorimeter Software

#

Imaging colorimeter companion software typically encompasses a complete processing chain from image capture to result output. A typical software architecture includes the following functional layers:

1.1 Image Acquisition and Calibration Layer

#

• Multi-exposure Modes: Supports various acquisition algorithms like SinglePic, multi-frame average denoising (MultiPic), and High Dynamic Range synthesis (HighDyn) to adapt to different luminance ranges and SNR requirements.
• Calibration Management: Manages and switches between different photometric/chromatic calibration files (e.g., Standard Illuminant A, four-color calibration), ensuring raw image data is correctly converted into physical quantities (cd/m², x/y coordinates).
• Geometric Correction: Corrects geometric errors such as lens distortion and image rotation.

1.2 Measurement Area Definition Layer

#

• ROI (Region of Interest) Definition: Defines the target areas for measurement and analysis within the image.
• Auto-alignment: Automatically identifies display borders or fiducial marks to compensate for DUT positional shifts during batch testing.

1.3 Analysis Algorithm Layer

#

• Statistical Analysis: Performs statistical calculations (mean, standard deviation, max/min, etc.) on luminance/chromaticity data within the ROI.
• Symbol Object Analysis: Independently identifies and evaluates discrete light-emitting elements (e.g., pixels, LEDs).
• Blob Analysis: Detects and quantifies anomalous regions (defects) in the image.
• Uniformity Analysis: Evaluates the spatial distribution uniformity of luminance and chromaticity.

1.4 Result Output Layer

#

• Pass/Fail Judgment: Compares measurement results with preset specifications to output a verdict.
• False Color Visualization: Maps luminance or chromaticity data to intuitive color images.
• Report Generation and Data Export: Generates test reports and exports data to Excel, CSV, or databases.

II. Region of Interest (ROI) Definition and Multi-Region Setting

#

2.1 Basic Concept of ROI

#

A Region of Interest (ROI) is the fundamental unit of image processing analysis. It defines which parts of the image the software “pays attention to” for calculation. Image areas not covered by an ROI are ignored and do not participate in statistics or analysis.

Functions of an ROI:

• Focusing Analysis: Eliminating interference from non-target areas like display borders or fixtures.
• Partitioned Evaluation: Dividing the display into multiple sub-regions to evaluate optical parameters for each separately.
• Defect Localization: Calibrating the spatial position of defects.

2.2 ROI Shape Types

#

Imaging colorimeter software typically supports various ROI shapes:

2.3 Applications of Multi-Region Settings

#

In practical inspection, multiple ROIs are often defined simultaneously. Typical configurations include:

Standard Partitioning for Uniformity Measurement:

Referencing VESA FPDM / SID IDMS standards, uniformity measurement typically involves dividing the display into equal partitions. For example, dividing the display area into 9 (3x3) or 25 (5x5) sub-regions, calculating the average luminance and chromaticity for each, and then evaluating uniformity using:

Luminance Uniformity = (L_min / L_max) × 100%

Nested ROI Layers:

In some applications, ROIs can be nested. For example:

• An outer ROI defines the entire active display area.
• Inner ROIs exclude non-active edge areas (some screens may have edge pixels that do not display normally).
• Further sub-regions for analysis are divided within the active area.

2.4 Auto-ROI and Template Matching

#

In automated production lines, fixed ROI coordinates may fail to align perfectly with the display area due to DUT positional shifts. Solutions include:

• Feature-based Auto-alignment: The software automatically identifies display borders (via light/dark edge detection) or preset fiducial marks to dynamically adjust ROI positions.
• Template Matching: Using a pre-saved standard image as a template, image registration algorithms calculate the offset and rotation between the current DUT and the template to compensate ROI coordinates.

III. Point of Interest (POI) Auto-Detection

#

3.1 Definition of POI

#

Point of Interest (POI) functionality is used to automatically search for and localize discrete light-emitting elements in the image, such as:

• Bright dot defects on a display.
• Individual LEDs in a keyboard backlight.
• Independent indicator lights on an automotive dashboard.
• Individual LED units in a MiniLED/MicroLED array.

3.2 POI Search Algorithms

#

POI search is typically based on the following steps:

1. Global or Local Threshold Segmentation: Dividing the image into “bright” and “dark” regions based on a luminance threshold. This threshold can be fixed or adaptive (e.g., proportional to neighborhood average luminance).
2. Connected Component Labeling: Performing analysis on segmented “bright” regions, grouping adjacent bright pixels into candidates.
3. Feature Filtering: Filtering candidates based on area, shape, and luminance to exclude noise or non-target regions.

3.3 Symbol Object Evaluation

#

In professional software (e.g., TechnoTeam’s LMK LabSoft), “Symbol Object” is a tool for precision evaluation of discrete elements. Unlike simple threshold segmentation, symbol evaluation features:

• Adaptive Edge Recognition: Using “Neighborhood” and “Foreground Range” parameters to accurately define symbol boundaries in fuzzy transitions. This does not rely on a single global threshold but rather on local adaptive judgment.
• Integral Luminous Flux Calculation: Even if symbol edges are blurred by optical diffusion, the software can collect all light emitted by the symbol through integration methods, accurately calculating its average luminance and total flux.
• Local Extremum Detection: Using adjustable circular “Point Sensors” (defined by Filter Size) to scan within the symbol area for local maxima and minima and their positions. This is useful for evaluating the luminance uniformity of individual elements.

Key parameters:

IV. Blob Analysis Algorithm: Connected Component Detection and Defect Quantification

#

4.1 Basic Concept of Blob Analysis

#

Blob Analysis is a classical algorithm in machine vision for detecting and quantifying “anomalous regions.” A “Blob” refers to a group of connected pixels sharing common characteristics (e.g., luminance, color). In display defect detection, Blob analysis identifies and quantifies defects such as:

• Bright Spots / Dead Spots
• Bright Lines / Dark Lines
• Local Color Spots
• Mura Regions

4.2 Blob Analysis Processing Workflow

#

Step 1: Image Pre-processing

Before Blob detection, images usually undergo:

• Background Modeling: Obtaining an “ideal background” (the expected luminance distribution without defects) via low-pass filtering or polynomial fitting.
• Difference Image: Subtracting the background model from the original image to get a “defect image,” where normal areas approach zero and defects appear as positive (bright) or negative (dark) values.

Step 2: Threshold Segmentation

Applying a threshold to the difference image marks pixels exceeding the threshold as “defect candidates.” Threshold setting directly impacts sensitivity:

• Too low: Noise fluctuations in normal areas are judged as defects (high false-positive rate).
• Too high: Subtle defects are missed (high false-negative rate).

Thresholds can be absolute (e.g., deviation > 2 cd/m²) or relative (e.g., deviation > 5% of background).

Step 3: Connected Component Analysis (CCA)

Performing labeling on the binary image after segmentation. CCA groups adjacent candidate pixels into a single Blob. Connectivity is commonly defined as:

• 4-connectivity: Neighbors in four directions (up, down, left, right) are connected.
• 8-connectivity: Neighbors in eight directions (horizontal, vertical, diagonal) are connected.

Step 4: Feature Extraction and Filtering

Calculating geometric and photometric features for each detected Blob and filtering based on preset criteria:

Step 5: Defect Classification and Judgment

Classifying each Blob into a defect type based on extracted features and grading severity:

• Critical: Defects severely impacting user experience (e.g., bright spots exceeding size limits).
• Major: Perceptible but less impactful defects.
• Minor: Tiny defects hardly noticeable under normal conditions.

V. Applications in Non-Rectangular Screen Inspection

#

5.1 Challenges in Non-Rectangular Inspection

#

Modern displays are no longer confined to standard rectangular shapes. Curved dashboards, circular smartwatches, and phone screens with cutouts (for cameras/sensors) pose new challenges for software:

• ROI Definition: Standard rectangular ROIs cannot precisely cover non-rectangular areas.
• Uniformity Evaluation: How to define reasonable partitions and evaluation standards in non-rectangular areas.
• Edge Processing: Complex edge shapes require special handling for edge-region Blob detection.

5.2 Solutions

#

Polygonal ROIs and Masks:

Software supports precise definition of active areas for irregular screens using polygonal or arbitrary masks. Masks can be generated by:

• Manually drawing boundaries.
• Importing screen contour coordinates from CAD drawings.
• Auto-detection: Utilizing light/dark boundaries in the image to automatically identify screen contours.

Adaptive Partitioning:

Traditional rectangular grids no longer apply to irregular areas. Software can employ:

• Non-uniform grid partitioning within the active mask area.
• Polar coordinate partitioning (for circular screens).
• Pixel-density-based adaptive partitioning.

Automotive Instrument Cluster Example:

A modern digital cluster may consist of multiple display areas of different shapes (e.g., two circular dials and a central rectangular information area). The inspection solution requires:

1. Defining independent ROIs for each area.
2. Executing independent uniformity analysis and defect detection within each ROI.
3. Consolidating results from all areas for a comprehensive verdict on the cluster.

VI. Generation and Application of False Color Maps

#

6.1 What is a False Color Map?

#

A luminance image captured by an imaging colorimeter is essentially a 2D grayscale matrix, where each pixel value represents luminance (cd/m²) or chromaticity. Human ability to distinguish grayscale differences is limited—typically only 20-30 levels. False color maps enhance data visualization by mapping these values to a color space (e.g., rainbow scale: blue -> green -> yellow -> red), transforming subtle differences invisible to the naked eye into obvious color contrasts.

6.2 Generation Process

#

1. Determine Mapping Range: Select the range of data to visualize (e.g., luminance 200-300 cd/m²).
2. Normalization: Linearly normalize data within that range to [0, 1].
3. Colormap: Map normalized values to RGB colors via a Lookup Table (LUT). Common schemes include:
   • Rainbow/Jet: Blue -> Cyan -> Green -> Yellow -> Red; high contrast but can be visually misleading.
   • Heatmaps/Inferno: Black -> Purple -> Red -> Yellow -> White; monotonically increasing luminance for better perceptual consistency.
   • Grayscale Enhancement: Maintaining grayscale while enhancing contrast.
4. Overlay Color Bar: Providing a correspondence table between colors and values alongside the image.

6.3 Value of False Color Maps

#

Luminance Uniformity Visualization:

For full-white measurement images, false color maps intuitively present spatial distribution. Engineers can immediately identify “dome” distributions (bright center, dark edges) or local anomalies.

Chromaticity Distribution Visualization:

Mapping colorimetric deviations (e.g., Δu’v’ relative to the center) to false color intuitively displays color uniformity, especially useful in backlight debugging.

Defect Labeling:

In result images, false color can overlay defect markers:

• Highlighting defect Blobs with red outlines.
• Using grayscale or low-contrast false color for the background.
• Labeling defect number, type, and area simultaneously.

Mura Visualization:

Mura defects often manifest as faint luminance variations that are invisible in raw grayscale images. Narrowing the false color mapping range (e.g., displaying only a +/-5% variation) makes Mura clearly visible, greatly improving analysis and debugging efficiency.

VII. Data Export and Statistical Analysis

#

7.1 Export Formats

#

Software typically supports several export formats:

7.2 Statistical Analysis Features

#

Built-in or external tool-compatible statistical features include:

Single-Device Statistics:

• Histogram: Frequency distribution of luminance/chromaticity values.
• Line Profile: Curves of luminance/chromaticity variation along a path.
• Scatter Plot: Distribution of color coordinates on a CIE chromaticity diagram.

Batch Data Statistics:

• Trend Chart: Trends of key parameters over time/batches.
• Cpk/Ppk Analysis: Process capability index evaluation.
• Control Chart: Monitoring if parameters exceed control limits.

7.3 Integration with MES/SPC Systems

#

In modern smart factories, inspection data is integrated with Manufacturing Execution Systems (MES) and Statistical Process Control (SPC) systems:

• Real-time Upload: Results for each DUT are uploaded to the MES via standard interfaces.
• Serial Number Binding: Data is bound to unique DUT serial numbers for full traceability.
• Anomalous Alarms: Automatic alarms trigger when consecutive defects or parameter drifts occur.
• Process Feedback: Data is fed back to upstream processes to support closed-loop quality control.

VIII. Summary

#

Image processing software is the core link that transforms photometric data into engineering value for imaging colorimeters. ROI definition determines “what to analyze,” Blob analysis and symbol evaluation determine “how to find defects,” false color mapping determines “how to present intuitively,” and data export/statistical analysis determines “how to form a quality loop.” In actual deployment, the completeness of software features, algorithmic flexibility, and integration capability with automated systems are often the decisive factors in whether an inspection system truly delivers value.

FAQ

#

Q1: How does Blob analysis work in display defect detection?

#

Blob analysis detects defects through a multi-step process: first, background modeling and image subtraction produce a defect image; then threshold segmentation marks defect candidate pixels; connected component analysis groups adjacent candidates into individual Blobs; finally, features like area, centroid, major/minor axis, and luminance deviation are extracted for filtering and classification. Blobs are categorized into defect types such as bright spots, dark spots, bright lines, dark lines, color spots, and Mura, with severity grades of Critical, Major, and Minor.

Q2: What is the role of false color maps in display inspection?

#

False color maps convert subtle luminance or chromaticity differences—invisible to the human eye—into obvious color contrasts, greatly enhancing data visualization. In display inspection, they intuitively present luminance uniformity distribution, chromaticity deviation patterns, defect annotations, and Mura visualization. For faint Mura defects, narrowing the color mapping range (e.g., showing only +/-5% variation) makes otherwise invisible patterns clearly discernible in the image.

Q3: How are non-rectangular screens (circular watches, automotive dashboards) inspected?

#

Non-rectangular screen inspection uses polygonal ROIs or arbitrary-shape masks to precisely define active display areas. Masks can be generated by manual drawing, CAD import, or automatic boundary detection. For uniformity evaluation, polar coordinate partitioning (for circular screens) or non-uniform grid partitioning replaces traditional rectangular grids. For multi-region composite screens like automotive clusters, independent ROIs are defined for each display area with separate analysis, then results are consolidated for a comprehensive verdict.

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