/
How to obtain image information capacity

How to obtain image information capacity

Information Capacity is a new measurement (introduced in 2024 for slanted edges) intended to predict the performance of Machine Vision/Artificial Intelligence (MV/AI) systems. Part of Imatest’s Image Information Metrics, Information Capacity is derived from MTF and noise measurements made at the same location; however, it is a much better predictor than either taken by itself.

To obtain the information capacity with any slanted edge module, the edge contrast should be 10:1 or less, with 4:1 (the ISO 12233 2014+ standard) strongly recommended. Results are more reliable with minimally processed images, converted from raw, but they are still useful for JPEG images, which have a type of non-uniform image processing called bilateral filtering, that sharpens the image near distinct edges but smooths is in relatively flat areas. 

To turn on the information capacity calculation

In the setup windows for auto ROI detection modules (SFRplus, eSFR ISO, Checkerboard, or SFRreg), make the selection in the Information capacity dropdown menu. We generally recommend Information cap - Auto edge noise detect, but two other selections may be slightly more robust if you know the type of image processing that has been applied: Info cap - mean noise for minimal or uniform processing or Info cap - smoothed peak noise for bilateral-filtered images (most in-camera JPEGs).  

image-20250627-214038.png
Information capacity dropdown menu

The same dropdown menu is available in the More settings window. It is also available in the SFR settings window.

Information capacity in bits per pixel is displayed in several places, but the most prominent is in the Edge/SFR plot:

image-20250627-215644.png
Portion of Edge/MTF plot, showing information capacities C4 and Cmax

  • C4 is the information capacity for signals with a 4:1 contrast range (the same as the chart contrst ratio)

  • Cmax is the maximum information capacity for the system, assuming it can reproduce the full range of digital numbers allowed by the the image file bit depth (a maximum of 255 for bit depth = 8, 65535 for bit depth = 16, etc.).

See Also