 |
| 1: Microlenses 2: X-Trans Color Filter 3: L/R Light Interception Filter 4: Phase Detection Sensor / Green Filter Pixel 5: Photodiode |
The 1/1.5-inch sensor has BSI pixels and integrates phase detection pixels for faster AF. Fijufilm says that an optical LPF is not necessary for its X-Trans sensors (both 1/1.5-inch and APS-C ones):
"X-Trans CMOS II adopts an original color filter array with a highly random pattern, eliminating the need for an optical low-pass filter (OLPF), which is used in conventional systems to inhibit moiré at the expense of resolution. This lets the sensor directly capture the high resolution performance of the FUJIFILM X20 lens. Also exploiting the enhanced performance of the EXR Processor II, Lens Modulation Optimizer (LMO) factors are calculated to compensate for aberrations and diffraction blur that occur when light passes through the lens and then are applied to produce images with amazing sharpness."
The noise of the new sensor is said to be reduced by 30%, improving the low light SNR.
Also, Nikon just announced two new mirror-less cameras with 1" sensors, the S1 and J3, which do not have IR-cut filters.
ReplyDeleteNikon also claims the filters are removed to provide sharper images. It's not clear if they are doing something extra in the image processing to remove the moire pattern, or if they depend on the user to remove it in Light Room or the Nikon image processing software
http://nikonrumors.com/2013/01/09/the-new-nikon-1-j3-and-s1-cameras-do-not-have-an-anti-aliasing-aa-filter.aspx/
-DP
Indeed, the official Nikon J3/S1 pages too say there is no OLPF:
Deletehttp://imaging.nikon.com/lineup/acil/bodies/j3/features01.htm
http://imaging.nikon.com/lineup/acil/bodies/s1/features01.htm
Someone wrote me recently to ask my opinion of the new Fuji "pattern".
ReplyDeleteThey said "The results so far all seem to suffer from various issues, smearing, zipper patterns, some sort of moire and coloured "pepper dots".
and asked me to comment on that.
My only comment is that Bayer is well studied and ISP for new patterns is always a challenge (Esp if they are sort of random!). Maybe someone else here has more insight to specifics?
Well, I think the white strip at the lower part of the eye looks somewhat unexpected.
Delete--
Best,
Iliah
Amazingly, they managed to capture both pictures without hardly any difference at all! Well, this is just a simulation, or worse, a photoshop trick.
ReplyDeleteXtrans is a nice CFA that can yield very good results with proper demosaicking. Problems such as smearing, zippering and pepper dots are the results of poor demosaicking while moire can be attributed to sharp optics. Even though Xtrans is more resistant to aliasing than Bayer, it would do well with a (weak) OLPF.
ReplyDeleteRandomness isn't harmful for CFAs, not being "well mixed" is. If a CFA has large neighborhoods without a color, then you have a problem. The missing color has to be interpolated over "long" distances and this amplifies both noise and artifacts. An example random CFA is demonstrated at http://imagealgorithmics.com/RGB_CFAs_Tested.html#RandomRGB
Finally, we are trying to meet the ISP challenge for new (and existing) patterns. We are in the process of commercializing a universal demosaicker that yields higher PSNR than all demosaickers on all CFAs.
Tripurari Singh: has imagealgorithmics used their algorithm on XTrans CFA? Will you publish he results?
ReplyDeleteThat would be a selling point for Fujifilm (or others) licensing your software especially with push from consumers :-)