Using "Shaman" filters for fine image correction
 |
At the left, you see the source picture - video capture taken from the regular VHS tape. At the right, you see the result achieved after three runs of "Convolution shaman" at different settings - no masking, no tricks at the host side, just one filter. And if you can't see these ugly color artifacts at the neck and forehead areas (left image), you should either adjust your monitor or keep out of the graphics area ;-)
|
 |
Step 1: smoothing the noise out
 Source video capture |
Our primary target will be these irregular color artifacts placed all over the face area. We are going to smooth them, but we are not going to smooth main image details like eyes, nose etc. So, launch "Convolution shaman", choose "Blur" and gently tweak "Soft threshold" until you get the result similar to the image at the right.
The idea behind "Soft threshold" is that filter automatically detects the "sharp" areas and tries to smooth the ones that are smooth enough already. This way it smooths forehead and neck areas without destroying eyes, nose and eyebrows, which are critical to face recognition.
|
 Smoothed result (blur radius 2 pixels) |
Step 2: sharpening main details
 Smoothed image from the step 1 |
Now after smoothing unwanted artifacts we try to sharpen the details we want to be clear. Use "Sharpen" or "Sharpen edges" functions of "Convolution shaman" for that.
The difference between "Sharpen" and "Sharpen edges" is that the latter algorithm sharpens mostly the edge area, that is, the areas of significant details. This way, again, we do not sharpen minor noise and artifacts but main details only.
|
 Sharpen edges (radius 1 pixel) |
Step 3: the final touch
 Sharpened image from the step 2 |
Now the picture looks almost perfect. Almost - because if you look close, you'll see the rest of these color artifacts - now they are smoothed and less pronounced, but still noticeable. And here we do one final touch, requiring the "Convolution shaman" full (shareware) version. We perform a wider blur in "Chromaticity only" apply mode (not present in free version), using "Soft threshold" as needed.
What's so great about these "Lightness only" and "Chromaticity only" apply modes present in full version? Why the real pro will, no doubt, need it? This requires a bit more explanation...
|
 Blur (radius 4 pixels, "Chromaticity only" apply mode) |
Another approach to step 1: let's sort them out!
 Same old source image |
There is a more "aggressive" way to eliminate the small artifacts: using the "Sorting shaman" filter's "Despecle" functionality. It's not as fast as "Convolution shaman" but in many cases is way more effective. Again, "Soft threshold" is the key to success...
"Chromaticity only" apply mode present in full version work especially well in such a cases...
|
 Despecle with "Soft threshold" engaged |
Bottom line: Web issues
One final note for a web designers: delicate smoothing the image with "Convolution shaman" or "Sorting shaman" as shown above have one positive side effect: since we eliminate minor noise and make picture smoother, it becomes much more compressible into, say, JPEG. For example, source image above is 9.1 kb, while despecled one is only 7.3 kb compressed at the same settings. Don't you want to squeeze some 20% out of you page weight, still maintaining (or even improving) the quality?
|