Skip to main content
Español

How to Calibrate to Match your Own Color Groups

  • Updated

When working with your own predefined color scale (for example, an 8-color lemon scale), it is important to calibrate the Computer Vision system to match your existing commercial classification.

This article explains the calibration methodology and how to align system color percentage ranges with customer-defined color groups.

 

Objective

The goal of color calibration is to:

  • Align Computer Vision color percentage measurements with your own existing color scale
  • Ensure inspection results match commercial expectations
  • Reduce discrepancies between visual reference charts and digital measurements

The calibration process allows us to translate a visual color scale (e.g., 4 lemon color stages) into precise percentage-based color ranges within the system. 

Calibration Methodology

To calibrate the system properly, we conduct structured test inspections.

Step 1: Create the Calibration Process

  1. Go to the Specification (Spec) section in the system.
  2. Select the relevant produce you want to calibrate.
  3. Create a new process and name it clearly (e.g., Lemon – Color Calibration). 
  4. Add the following screens to this process:
    • Inspection Details
    • Manual Attributes (for inspector’s manual color assignment)
    • Photos
    • CV Attributes

This setup allows both automatic (CV) and manual color data capture for comparison and validation. 

Step 2: Take Calibration Inspections with Manual and Automatic Data

To generate reliable color data from your system:

  1. Conduct structured calibration inspections for each customer-defined color group. 
  2. For a typical 4-group (4-color) scale: 
    • Create 10 inspection sessions per group.
    • In each inspection, include 5–10 fruits photographed together (not one by one). 
    • Capture 3 photos from different angles showing all fruits in that inspection.
  3. During each inspection, record:
    • The Computer Vision (CV) color measurements (Color Percentage)
    • The manual inspector’s color group attribution via the manual attributes screen
      Combining manual and automatic data helps ensure that the system measures colors in alignment with human assessment. 

Result:

  • 40 inspections total (4 groups × 10)
  • At least 200 fruits sampled (4 groups × 10 × 5 fruits)
  • 3 images per inspection for robust data. 

Step 3: Fetch Automatic Color Results and Identify Group Limits

Once the inspections are completed:

  1. Run or fetch the color percentage report that the CV system generated for each inspection. 
  2. For every color group, organize the data into a distribution of measured color percentages (e.g., min, max, average). 
  3. Analyze this distribution for each group to determine where natural boundaries lie between stages. 
    • You’ll see how the measured percentages cluster for each group
    • This insight helps you define thresholds that separate one color group from the next. 

Step 4: Configure Color Groups by Calculated Limits

Now use the results from Step 3 to configure the color groups:

  1. Define each group’s name (matching your customer scale) and assign the corresponding percentage ranges discovered in Step 3. 
  2. These configured ranges tell the system how to assign a CV percentage to a color group. 

Goal:

  • Minimize misclassification
  • Align system measurements with your defined visual color stages
  • Reduce overlap between adjacent groups.

Step 5: Test with New Fruits and Verify Results

Finally, validate your calibration:

  1. Conduct new inspection sessions on fruits not used during calibration.
  2. Capture both automatic CV data and manual inspector assessments again.
  3. Compare the new automatic classifications with manual assessments. 
    • Verify that the calibrated percentage ranges classify fruits in a way that matches visual expectations
    • If significant mismatches are found, consider refining the boundaries and repeating tests. 

Summary

Color calibration ensures that Computer Vision measurements match the customer’s established visual scale.

By:

  • Creating structured inspections
  • Collecting a minimum of about 120 representative images
  • Analyzing the percentage distribution per group
  • Configuring manual percentage ranges

We achieve accurate, objective, and commercially aligned color classification.

 

 

Related articles