> > yuvdenoise has some form of sharpening available with the -S > > parameter, you might try that. > > I did, and I believe that the unsharp mask is more efficient. From what I > understand it is more or less a way or improving the ratio noise / signal. > But I might be wrong on this one...
yuvdenoise's main purpose is to temporally lowpass filter, reducing temporal noise and increasing SNR. Since motion is a joint time-space effect, temporal filtering often softens the video also. I think sharpening is in yuvdenoise to try to counter this. > > In general, the problem with > > sharpening video is that it increases the high frequencies (edges) and > > in doing so raises the noise level. > > See .... :-) I am wrong.... However, if the noise is increased how come that > the picture looks better ? Well, better is somewhat subjective, but sharpening enhances edges (increases the contrast) and makes the video look, well, much sharper, more in-focus. It's always done in digital still photography where size isn't such an issue. But most of the image signal energy lies at lower frequencies, while the noise is typically evenly distributed across frequencies (depends on the noise source, of course). Since sharpening increases high frequencies, the effect is to increase the noise more than the signal. It's a matter of degree - if you have clean sources, sharpen away! If they're noisy, then denoise first, or limit the degree of sharpening based on visual preference and bitrate. > Yes... but again with my little knowledge, I understand it is based > on geometry rather than anything else. So applying it to YUV or RGB > space colour will not change anything. We are not in the "same > space" are we ? If you do yuv->sharpen directly on each of Y, U, and V, you will get different results than yuv->rgb->sharpen->yuv. Neither the conversions nor the sharpening are linear. There would probably be no reason to sharpen the chroma (it's subsampled, so it doesn't have the higher frequencies anyway), and I think you could just apply a monochrome sharpening algorithm to the luma. It would be nice to have a Y-only y4msharpen, and it would not be hard to do. I'm not volunteering right now though. What I'd really like is a general spatial filtering routine for yuv streams. > See above for my very poor excuse not having the data... I still > believe it is worth doing it, can't we get rid of high frequency by > using - N after ? Would it defeat the purpose ? -N doesn't filter out high frequencies, it just encodes them with fewer bits. This because most of the image is low frequencies. Taken to extremes, the result is a loss of detail and an increase in high-frequency (quantization) noise. Again, it's going to be subjective. If it did remove high frequencies, then it would defeat the purpose. There are at least two sharpening methods I know of: one is using a linear spatial filter that directly amplifies high spatial frequencies. You mentioned unsharp masking, which is a nonlinear filter that does increase high frequencies as a side effect but probably increases the noise less. Following either with a lowpass filter would just soften the picture again. Sorry if I'm losing the non-engineers... Dan ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Mjpeg-users mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/mjpeg-users
