.Currently implemented image operations: Load, Set, Copy, Brightness, Stretch, Blur, Invert, Threshold, Shrink, Expand, Edge, Salt, Pepper, Noise, Noize.
I implemented Average, but my test program doesn't support enough images to actually test it.
In section 2.6, Davies talks about his time weighted averaging technique:
"Although the learning routine used here has not been found in the literature, it has been used in the author's laboratory, and was probably developed independently elsewhere."
I think this algorithm can be expressed as a set of 1D Kalman filter estimating the grayscale value of each pixel. The effect would be the same, although the scaling factors would be different. If the image is captured from a CCD, each observation value could be given an appropriate observation error. I'm not sure what the action model used to increase uncertainty over time should be. I suspect it would be a tuned parameter.
No comments:
Post a Comment