> softens the picture a lot. Median filtering in general is great for > noise distributions with long tails (non-gaussian, impulsive noise) > since it can exclude large outliers rather than averaging them in like > linear filtering, but the flip-side seems to be that edge detail gets > obliterated. There are lots of different median-filtering schemes out
It does soften it of course but I guess "a lot" is subjective. Seemed strange though that the documentation mentions 'sharpening' as a side effect. My statement about obliterating edges is only partly correct. My tests do show a noticeable loss of detail. However, one of the hallmarks of median filtering is actually preservation of detail. Confused? I was too, until I actually traced the code. It doesn't actually do median filtering; it computes what it calls a pseudo-median when enough surrounding pixels are within the given thresholds, and failing that it uses a regular mean. It's probably close to a true median filter for some inputs (mainly those with low or sporadic noise). I'm sure it's also a lot faster, since a median filter is a simple concept (order the surrounding pixel values and take the median) that isn't so simple to code up. I was using the filter on high-noise inputs with too low of a threshold, and so it resorted to averaging most of the time. Previously I had used it after yuvdenoise had removed most of the gaussian noise, and that worked out a lot better. I may hunt around for a true median filter code, just to see what that might do. Dan ------------------------------------------------------- This SF.net email is sponsored by: Etnus, makers of TotalView, The debugger for complex code. Debugging C/C++ programs can leave you feeling lost and disoriented. TotalView can help you find your way. Available on major UNIX and Linux platforms. Try it free. www.etnus.com _______________________________________________ Mjpeg-users mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/mjpeg-users
