Why the feature should be RST invarient..when extracting a feature from an image..

A good feature should have high varinace and it should be RST invariant for image. Why is it so?

[Walter Roberson]

This appears to be a theory question rather than a MATLAB question.

Answer 1

[Walter Roberson]

If "RST" here stands for "Rotation, Scaling, Translation", then under what circumstances would it be beneficial to have a feature that was varient under one of those operations?

[Image Analyst]

One example, robots to "pick and place" parts from a bin onto an assembly they're building.

[Walter Roberson]

True, size (scale) could be important in some contexts that involved calibrated sensors.

Answer 2

[Image Analyst]

That's only true for certain situations. I'm sure you can imaging situations were you'd like them to vary and situations where you'd like them not to vary. Though I can't think of any situations where you'd want the feature measurements to have a high variance instead of a low variance. Invariant and variance are different concepts, but not antonymns (opposites) - consult a dictionary.