Am I oversharpening
in [Digital Cameras]
|
From: JPS on 6 Jun 2005 23:19 In message <1118002917.281821.236530(a)g44g2000cwa.googlegroups.com>, george_preddy(a)yahoo.com wrote: > > >JPS(a)no.komm wrote: >> >Interpolation never adds optical resolution. >> > >> >Try it for yourself, download a satellite picture of the world and >> >interpolatively upscale it using Photshop. Make it a bigger and bigger >> >image until you can actually see yourself smiling in the picture. >> >> Don't be ridiculous. Everyone arguing with you knows that you can't get >> more detail by upscaling a bitmap. > >Each Bayer expoure is a bitmap, interpolated values are inserted >between optical data points in each channel to upscale each RGB >exposure to the sensor's overall monochrome dimension. This is >upscaling. No, it is not. There is a witness at every 2D location in a Bayer capture. Each one captures some information that would not have been captured if each arbitrary group of RGGB was in the same 2D location. >That is what interpolation means afterall, insterting a guess between >data points. There is no difference if you do it on a desktop or in a >camera, as my posted full size samples showed so nicely. Your posted, full sized images show aliasing (edges and textures look fake, with a "snap to grid" effect), and poor color rendition. >There is virtually no loss in full color resolution when you downsize a >Bayer to 25% of its recorded size, because it was already digitally >upscaled 400% as recorded. That's not what really happens, at all, in the luminance channel. It's true in the hue/saturation channels in an HSL model, or a/b in a Lab model, but that is true of Foveon images as well, unless the subject is a luminance landscape blue/red resolution test. >> The luminance, the most significant part of the capture, is not upscaled >> at all. Red and blue resolution are upscaled to 200%, and green >> resolution is upscaled 70.7%. >Red and Blue are upscaled 400%, Green is upscaled only 200% but is >effectively reduced by half in order to topologically align the RGB >composite image. That is when viewing area, but area is not the default method of referring to resolution. >As a result, Green accuracy is somewhat imporoved, >but the resulting interpolated Green quantity is utterly identical to >Red and Blue. Obviously, since every final pixel has an R, G, and B >cahnnel. Obviously, you're making this up. -- <>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<> John P Sheehy <JPS(a)no.komm> ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
From: JPS on 7 Jun 2005 00:07 In message <1118012948.485965.226670(a)z14g2000cwz.googlegroups.com>, george_preddy(a)yahoo.com wrote: > > >Bart van der Wolf wrote: >> <george_preddy(a)yahoo.com> wrote in message >> news:1117934606.605577.297030(a)g47g2000cwa.googlegroups.com... >> SNIP >> > Bayer DSLRs are characteristically blurry and soft because the true >> > optical resolution is the resolution of each, individual, tiny, >> > subdivided, R, G, or B exposure, [...] >> >> How typical, only half-truths (if any). >> Did you forget about lens quality and the anti-aliasing filter, which >> are the true source for blur? Resolution is limited by that, the >> sampling density, and output magnification. > >A camera can't have an a blur filter and have high optical quality. You use the term "blur filter" because it sounds so negative, but anti-aliasing spreads around data in a manner that could only be considered a tiny blur, if you're going to call it a blur. It does not reduce much detail to an irretrievable level. >Intentionaly introducing blur and optical quality are not compatible >notions. It is absolutely *necessary* to use an AA filter, to have have points of light from the subject properly located in the final bitmap. If you created a time-lapse animation of a star trail with both an SD9 and a bayer DSLR with a proper AA filter at the same time, the Bayer star trail animation would move smoothly across the image, and have consistent illumination. The SD9 animation would jump and blink, because an unaliased, under-sampled image can not precisely locate a point of light. >Granted, blur is the lesser of two evils--vice having color >moire rampant in any area of detail. Sampling in a bayer pattern requires a stronger AA filter, to reduce color moire, but no system can get away with not having an AA filter, unless the resolution of the lens falls way below the pixel pitch, and that simply is not the case with The Sigma SD cameras. They have a large pixel pitch, and many lenses outresolve them by a wide margin, resulting in aliasing. >"Lesser of two evils" is hardly an endorsement for using mosiacs to >keep a camera's optics super cheap. Hardly; the fake-texture, snap-to-grid look of the SD9, apparent to anyone with a properly functioning visual system and right hemisphere is an evil, if you want to see the true texture of the subject. -- <>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<> John P Sheehy <JPS(a)no.komm> ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
From: JPS on 7 Jun 2005 00:11 In message <cFFkNeAKN$oCFwh2(a)objectech.co.uk>, Ken Tough <ken(a)objectech.co.uk> wrote: >So even your foveon-based camera has an AA filter. No, it doesn't. The SD10 has microlenses, to collect the equivalent of a near-100% box filter (and increase sensitivity), but the SD9 has no microlenses and no AA filter at all, and captures light from only 30% of the surface. The diameter of the photosensors in the SD9 is only 55% of the spacing between pixels. -- <>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<> John P Sheehy <JPS(a)no.komm> ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
From: JPS on 7 Jun 2005 00:24 In message <1118035981.973077.252870(a)g47g2000cwa.googlegroups.com>, george_preddy(a)yahoo.com wrote: >Tetractys wrote: >> George Preddy wrote: >> > In general, if you are sharpening at all then >> > you are oversharpening. >> This is an untrue statement. Sharpening is part >> of any proper digital workflow. Look here: >> http://www.luminous-landscape.com/techniques/process.shtml >Only Bayer workflow needs artificial sharpening. That is one of >several reasons why "digital" (read: Bayer) images are always paper >flat. They're not "paper-flat" if you downsize them to 2268*1512 with point sampling (nearest neighbor algorithm), but then they develop other problems, just like the SDx images. -- <>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<> John P Sheehy <JPS(a)no.komm> ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
From: Ken Tough on 7 Jun 2005 01:38
JPS(a)no.komm wrote: >Ken Tough <ken(a)objectech.co.uk> wrote: >>So even your foveon-based camera has an AA filter. >No, it doesn't. The SD10 has microlenses, to collect the equivalent of >a near-100% box filter (and increase sensitivity), but the SD9 has no >microlenses and no AA filter at all, and captures light from only 30% of >the surface. The diameter of the photosensors in the SD9 is only 55% of >the spacing between pixels. It isn't an issue of how much of the surface is sampled, it's the sampling itself that's at issue. If it doesn't have an AA filter, it needs it. (The microlens is the equivalent of an integrate-capture-hold on audio sampling). -- Ken Tough |