Stray Light Masking — Imatest Documentation

Stray light (flare) documentation pages

Introduction: Intro to stray light testing and normalized stray light | Outputs from Imatest stray light analysis | History

Instructions: High-level Imatest analysis instructions (Master and IT) | Computing normalized stray light with Imatest | Motorized Gimbal instructions

Settings: Settings list and INI keys/values | Standards and Recommendations | Configuration file input

Page Contents

This page describes Imatest’s light source Mask Methods.

Masking (shorthand for “light source masking”) is the process of creating a mask to exclude the direct image of the light source from the analysis. The direct image of the source is the small region in the image that represents the true size of the light source (i.e., if there were no stray light or blooming in the image). The direct image of the light source is not stray light, which is why it should be masked (ignored) in a stray light metric image. For reference images, the term "detection" is used instead of masking.

Mask Methods

The Imatest stray light source analysis has several options or methods available for masking the light source in the image(s). Different mask options have different inputs, allowing the user to configure the masking process and accuracy.

Method	Description	Analysis Image Mask	Reference Image Detection
Geometric Camera Model + Circle	Use a geometric camera model and the angles of the capture to locate the center. Apply a circular mask.	Yes	(Via Same As Analysis or Manual Circle)
Image Processing	Use image processing methods to create the mask.	Yes	Yes
Image Processing + Circle	Use the centroid of an image processing mask as the center. Apply a circular mask.	Yes	Yes
Manual Circle	Specify the center of a circular mask.	No	Yes
None	Do not mask.	Yes	No
Same As Analysis	Use the analysis image mask settings for reference image detection.	No	Yes

Geometric Camera Model + Circle

Description

The Geometric Camera Model + Circle method uses a geometric camera model to project the source angles to create an image point. A circle of a specified radius is placed at the center of the projected point and is used as the mask for the image. This method can avoid over-masking in the presence of large areas of saturation.

Inputs

- Geometric camera model
  - Geometric Camera Model: The camera model type and parameters
  - Negate Azimuth: The option to negate the field angle
  - Negate Field Angle: The option to negate the field angle
- Circular Mask
  - Circle options (radius only)

Mask Calculation

1. Negate the field and azimuth, as specified
2. Compute the camera-relative world point corresponding to the light source:
3. Project the camera-relative world point through the geometric camera model
4. Create a circular mask centered on the computed image point

Assumptions

- There is only a single light source in the image
- The user-provided geometric camera model well describes the camera
- The user-provided source angles well describe the angles of the source relative to the camera
- The user-provided source angles are in degrees
- The circular portion of the mask assumes relatively low distortion
  - The light source size is constant over the images
  - The image of the light source is always a circle

Notes

- The Geometric Camera Model + Circle method is only directly available for image analysis masking
- To use this method for reference image detection, either:
  - Use the Manual Circle method, where the manually selected point is the principal point
  - Use the Same As Analysis method. IN the case, the field angle and azimuth angle are assumed to be 0° for the reference image
- The maximum field angle currently allowed is 90°
- Other geometric camera models may be available in future versions of Imatest
- The field and/or azimuth angles may need to be negated to get the mask to align with the image of the source
- Imatest internally uses an IEEE Std. 2020 [1] Type IV image coordinate system
- The analysis image(s) must have a defined correspondence to the source angle via either:
  - Image file(s) with source angles in the filename and corresponding metadata settings
  - A stray light configuration file with angles defined

Image Processing

Description

This method uses image processing-based methods to mask out the light source in the image.

Inputs

- Level Threshold Type: The type of threshold used for localizing the light source in the image.
- Level Threshold (Value or Percent): The threshold above which is considered the image of the light source, as either a value (digital number) or a percentage of the maximum value in the image.
- Minimum Source Level: The minimum image value (digital number) that can be considered a direct image of the light source. This is used to prevent masking stray light when the source is out of the FOV.
- Mask Area Threshold: The minimum number of pixels that the direct image of the light source should be (minimum area in pixels for the mask). This setting is meant to prevent noise or other small bright artifacts from being detected as potential mask regions.
- Image Close Radius: This setting is only accessible in the INI. The radius of an image close morphology step during Image Processing. This helps to remove holes in the mask. Set to 0 to skip this step.
- Multiple CC Strategy: This setting is only accessible in the INI. The strategy for how to deal with multiple connected components when computing the mask.
- The minimum pixel count threshold

Calculation

Calculation Overview

1. Compute the level threshold for the image or use the entered Level Threshold (Value)
2. Check that the level threshold for the image is greater than the entered Minimum Source Level. If less, the source is assumed to be outside the FOV and no mask is generated for the image.
3. Create an initial mask by finding all of the pixels greater than the level threshold
4. (Optional) Perform an image-closing morphology with a disk of the user-specified size (Image Close Radius). This helps to fill in small holes within the mask
5. Compute the number of connected components in the mask
  - If zero connected components are found, then the light source is assumed to be out of the FOV
6. Filter out any connected components that have an area smaller than the user-provided Mask Area Threshold
  - If zero connected components remain, then the light source is assumed to be out of the FOV
7. Apply the strategy for selecting from one or more connected-component (Multiple CC Strategy)

Level threshold calculation

If the Level Threshold Type is Percent of Max:

1. Calculate the maximum value in the image
2. The threshold is the user-provided percentage (Level Threshold (Percent)) of this value

If the Level Threshold Type is Value:

1. Use the user-provided threshold (Level Threshold (Value))

If the Level Threshold Type is Otsu’s method

1. Use Otsu’s method to calculate a threshold from the image

Multiple Connected Component Strategies

If there is only one connected component:

- This is assumed to be the light source.
- A mask of just this connected component is returned

If there are no connected components:

- The light source is assumed to be outside the field of view.
- A mask with no light source(s) is returned

If the user-provided multiple-connected component strategy (Multiple CC Strategy) is “Don’t mask”:

- A mask with no light source(s) is returned

If the user-provided multiple-connected component strategy (Multiple CC Strategy) is “Pick Biggest”:

- The largest of the remaining connected components is assumed to be the light source
- A mask of the largest connected component is returned
- This option is useful when there are small ghosts/noise that are being confused with the direct image of the source.

If the user-provided multiple-connected component strategy (Multiple CC Strategy) is “Pick Smallest”:

- The smallest of the remaining connected components is assumed to be the light source
- A mask of the smallest connected component is returned
- This option is useful when the direct image of the source is small (a few pixels)

If the user-provided multiple-connected component strategy (Multiple CC Strategy) is “Keep All”:

- The union of all connected components will be used
- This option is useful for debugging masking settings

Assumptions

- There is only a single light source in the image

Notes

- The Image Processing method may not provide accurate localization of the direct image of the source if the level light source is above saturation and significantly blooming. The Image Processing + Circle method is recommended in the case where the light source is saturated and significantly blooming.

Recommendations

- Level Threshold (Value)
  - Pick the largest value (digital number) that the direct image of the source is always greater than in the image.
- Level Threshold (Percent)
  - If the direct image of the source is completely saturated, set it to 100%
    - If saturated, it is recommended to use the Image Processing + Circle method rather than just Image Processing because blooming and stray light may cause the Image Processing method to mask out more pixels than should be, leading to an underestimation of the stray light
  - If the Mask Method is Image Processing, then the goal of the Level Threshold is to mask only the direct image of the source. Using a value between 95% and 100% will work well in most cases (a hot pixel being an exception)
  - If the Mask Method is Image Processing + Circle, then the goal of the Level Threshold is to get a good estimate of the center of the direct image of the source, so that a circle of a specified radius can be drawn there. Using a value between 90% and 95% will work well in many cases
  - If following IEEE-P2020 Pre-Release, set to 95%. Note that this may not work well for images that exhibit saturation and/or blooming.
- Minimum Source Level
  - When configuring Test Image Mask Options, pick a value between the dimmest value (digital number) of the direct image of the source when it is in the FOV and the brightest stray light value in the image when the source is outside the FOV.
  - When configuring Reference Image Mask Options, set the value to 0 (the source should always be in the reference image)
  - Note: This setting is only useful if the level threshold type is Percent of Max or Otsu’s Method
- Mask Area Threshold
  - Use the default. Larger values will cause larger masses detected by the Level Threshold to be ignored.
- Image Close Radius (INI Only: image_close_radius)
  - Use the default. Larger values will fill in bigger holes in the mask.
- Multiple CC Strategy (INI only: multiple_cc_strategy)
  - Use the default.

Image Processing + Circle

Description

The Image Processing + Circle radius method uses two “steps” to determine the masked region. In the first step, the light source is localized via Image Processing methods. In the second step, a circle of a specified radius is placed at the center of the localized region and is used as the mask for the image. This method can avoid over-masking in the presence of large areas of saturation.

Inputs

- Step 1
  - Image Processing options
- Step 2
  - Circle options (radius only)

Mask Calculation

1. Apply the “first step” (Image Processing) to the image to determine a mask region
  1. If no mask region is found, the light source is assumed to be out of the FOV
  2. If a mask region is found, the light source is assumed to be in the FOV
2. Compute the centroid of the mask region
3. Apply the “second step” by creating a circle of specified radius at the found centroid to use as the final light source mask

Assumptions

- There is only a single light source in the image
- This method assumes relatively low distortion
  - The light source size is constant over the images
  - The image of the light source is always a circle

Notes

- This method does not perform well if the light source is partially in and partially out of the FOV.
- The centroid is assumed to be relatively stable in the presence of noise or saturation.
- The centroid may be a non-integer (sub-pixel) value. Therefore, the number of masked pixels is not constant.
- The first step is currently always an Image Processing mask method.

Manual Circle

Description

The Manual Circle method places a circular mask centered at a specified point.

Inputs

- Circular Mask
  - Circle options

Mask Calculation

- See Circle calculation steps

Assumptions

- There is only a single light source in the image
- The circular portion of the mask assumes relatively low distortion
  - The light source size is constant over the images
  - The image of the light source is always a circle

Notes

- Imatest internally uses an IEEE Std. 2020 [1] Type IV image coordinate system

None

Description

Do not mask out the source.

Inputs

N/A

Calculation

N/A

Assumptions

N/A

Notes

- If the source is in the FOV, this will count both the stray light and the direct image of the source as stray light.
- None masking cannot be used for a reference image-based normalization method (Direct Reference Image or Lambertian Reference Image).

Same as Analysis Image

Description

Use the same settings as the analysis image masking for the reference image detection.

Inputs

Set via reference image mask settings.

Calculation

The chosen analysis image mask method.

Assumptions

N/A

Notes

- This method only applies to the reference image masking.
- This can be used to use the Geometric Camera Model + Circle for the reference image detection. The reference image is assumed to have the source at azimuth = 0°, field angle = 0°.
- If selected, the analysis image mask method cannot be none.

Helper Methods

Circle

Description

The circle mask method places a circular disk at a specified position (which may not be within the image) with a given radius.

Inputs

- Center: the center image coordinate. Internally, an IEEE Std. 2020 [1] Type IV image coordinate system is used. Depending on the masking method, the center may be calculated from an upstream masking step or directly provided.
  - Image Center Type: The method for specifying the image center
    - Numeric Image Center: Use the numeric center of the image
    - User-Specified: Use a user-specified image coordinate
  - Image Coordinate System: The image coordinate system of the user-specified point (only used if the mask center type is User-Specified)
  - Mask Center: The user-specified location of where to center the mask (only used if the mask center type is User-Specified)
- Radius
  - Radius Method: The method for specifying a mask radius
    - User Specified: The radius in pixels
    - Focal Length Ratio: Use the ratios of the light source and camera to compute a radius
  - Mask Radius: The radius (in pixels) to use for a circle that would represent the direct image of the light source (Only used if the radius method is User Specified)
  - Camera Pixel Pitch: The pixel pitch of the camera (Only used if the radius method is Focal Length Ratio)
  - Camera Focal Length: The focal length of the camera (Only used if the radius method is Focal Length Ratio)
  - Source Focal Length: The focal length of the light source (Only used if the radius method is Focal Length Ratio)
  - Source Pinhole Diameter: The diameter of the light source (Only used if the radius method is Focal Length Ratio)

Radius Calculation

IEEE Std 2020:2024 [1] provides a method to compute the radius of a mask from properties of the light source and camera

where

: radius of mask in pixels

: pixel pitch of the camera in μm/px

: focal length of the camera in mm

: focal length of the collimator in mm

: diameter of the pinhole in the collimator in mm

Mask Calculation

1. Generate an x, y location for each pixel in the image. This is in the IEEE Std. 2020 Type IV [1] image coordinate system.
2. The mask is all pixels that satisfy

where

: radius of mask in pixels

: an image coordinate in the IEEE Std. 2020 [1] Type IV image coordinate system.

: the circle center in the IEEE Std. 2020 [1] Type IV image coordinate system.

Assumptions

- There is only a single light source in the image
- This method assumes relatively low distortion
  - The light source size is constant over the images (the radius does not change)
  - The image of the light source is always a circle (the shape of the mask does not change)

Notes

- This method is used for the Image Processing + Circle, Geometric Camera Model + Circle, and Manual Circle masking methods

Warnings

When the image of the light source is partially out of the FOV, the assumptions used to generate a mask may be violated.
- This may lead to improperly defined masks.
- It is recommended that all light source masks be checked.
When the image of the light source is entirely out of the FOV, the assumptions used to generate a mask may be violated.
- This may lead to improperly defined masks.
- It is recommended that all light source masks be checked.

Generic Notes

Mask creation channel
- If the analysis channel is Luminance or RGB + Luminance, then the luminance channel is used to create the mask.
- Otherwise, the mean of all provided channels is used to create the mask.
Mask application channel
- For each capture position, the same mask is applied to each of the analysis channels.
In color-mapped metric images, masked pixels are indicated in pink.

References

IEEE Standard for Automotive System Image Quality, IEEE Std 2020-2024, 2024, doi: 10.1109/IEEESTD.2024.10935839. ↩^a ↩^b ↩^c ↩^d ↩^e ↩^f ↩^g