TRV-950 stills .... 8x10 prints look great...a hypothesis on why.

[Mike Sanchez]

Punchline: 3CCD capture results in roughly 2X the resolution of a single CCD capture....and TRV-950 stills benefit from this. Details below....

I recently acquired a TRV-950. Prior to this acquisition my sources of image acquisition were excellent Canon bodies, lenses and mostly Fuji film along with a bit of video from an old Canon 8mm Camcorder. Used a film scanner to obtain digital images.

I went with 3CCD because color, reproduced with 3 CCD's, I hoped would be more accurate in reduced lighting and it is...but this is not my interesting issue.

I've read time and again about how poor stills are when using video cameras.....and......for the older single CCD 270,000 pixel cameras I would agree.

However, I have encountered something unexpected with the TRV-950 stills and I think that the results that Foveon is obtaining from their 3 layer capture explain what is happening.

With my, ostensibly, 1MP TRV-950 I am able, via bi-linear interpolation, to obtain really outstanding 8x10 prints at 250 dpi on thermal and inkjet printers. I have not tried larger yet, but, I suspect I can go a bit larger yet.

Why is this? Well, I think the answer resides in the recent series of reviews provided by dpreview.com on the Foveon sensor. The reviewer carefully examined the resolution of the Foveon sensor with the Canon 6.3MP single CCD.

Results indicated that when the image is formed from single pixel R, G, and B (like 3CCD or Foveon chip) the "effective" resolution (experimentally determined by comparing with the 6.3MP Canon single CCD) is roughly twice what the single CCD array is capable of with Beyer interpolation.

In other words, my TRV-950, at 1MP effective, is really 2MP effective because of the extra information associated with capturing all 3 RGB records completely rather than interpolating them from a 1MP array.

I find this a really fascinating and powerful benefit of the TRV-950 and unexpected as well.

Thought folks might be interested.

These results are partly mitigated in very low light without flash where, due to the CCD size and attendant lower light sensitivity, noise becomes an issue.

[Tom Hardwick]

I've made many A4 ( 8" x 11" approx) glossy prints from the TRV900 in progressive scan mode, and they look fine. OK, it's not the greatest resolution in the world and there's no lee-way for cropping, but when you have 12.5 pictures per second to choose from, the image is in the moment, and the moments in the image.

Few digital still cameras have a 12x zoom, and those that do certainly aren't f1.6 to f 2.8. The 900 offers a vast range of apertures and shutter speeds, control over white balance & focus. You have the choice of two viewfinders, one of which moves every which way. Stereo sound is recorded at the same time. A built in ND filter and three chips into the bargain.

tom.

[Ronald Lee]

HI Mike,

That is interesting. I assume that you are using photo mode right? Or are you taking frame grabes from your interlaced video?

Can you explain how you go about "bi-linear interpolation", I assume this is you deinterlacing it somehow? Perhaps deinterlacing the video first and then pulling the still, or deinterlacing it after, in photoshop?

Also, if you are using photo mode, doesn't it take a progressive still, so the deinterlacing is not needed? Mind you, I am not sure what "bi-linear interpolation" really is, I am only guessing it some way of smoothing out the image.

Thanks

[Mike Sanchez]

Ronald,

I am using Photo Mode (1152x864) gives a 2.85 MegaByte file.

This file has ~3 Million Pixels as its capture origin pixels compared to a single CCD with 1 Million Effective Pixels ~ 1 Million capture pixels. Both have a ~ 2.85 Megapixel file in RGB space, however, the TRV-950 has 3 times the actual data......the single CCD has already undergone Beyer interpolation to fill in the RGB points not captured in the Single CCD setup.

Bi-linear interpolation is one step up from linear interpolation just uses data on both sides (in 1-D) of the point of interest rather than forward or backward interpolation.

It is one of the default selections in Photoshop.

So, I interpolate the 2.85 Megapixel image (which has no interpolated pixles

[Chris Thiele]

Hi Mike,

I am a little confused, (not hard for me)

The TRV950 & PDX10 capture 1 mega pixel per chip making them a 3mg still camera.
Therefore they will be a lot better than a 1mg camera.
The file size is dependant on the compression used for each photo. Therefore file size will vary dending upon the amount of info it stores, eg a white wall will be a smaller size photo than a garden of flowers.
Therefore your camera is not a 1 mega pixel camera in photo mode but a genuine 3 meg captured in RGB (1 meg each)

[Mike Sanchez]

Chris,

Thanks for the interest and the reply.

Both the TRV-950, with its 3 million pixel R, G, B capture, and a 1 million pixel camera with a normal R, G, B color filter array placed on it will provide a 3 Megabyte file.

The TRV-950, because it captures at 3 million pixels for R, G, and B is not a 3 MegaPixel Effective (and Sony lists it as current "tradition" and "rules" require - as a 1.0 Million Effective Pixels). This is because the Luminance channel would be constructed from both the TRV-950 and the 1.0 Megapixel camera from 1 Million Pixels. Hence, if one converts the R, G, B signal for both cameras to uncompressed YCbCr the luminance information is the same for both.

However, the Cb, Cr channels are very, very different. In the case of the 1 Megapixel camera the uncompressed Cb, and Cr channels will have to be populated from R, G, B data that is incompletely captured for all pixels in the R, G, B array because of the fact that there is a color filter array sitting on top of a single CCD. Much of the color is interpolated (Beyer or some intelligent form of).

BUT....for the TRV-950.....the Cb, Cr channels are populated with data captured at the time of scence capture. This extra data does add some perception of detail, however, not to the luminance channel.....the the chroma channels. It is visually detectable and the author at dpreview.com has done a stupendously good job of comparing a single CCD visual resolution to the foveon (equivalent of 3CCD) capture and concluded that the 3CCD structure offers ~ twice the apparent visual resolution of the its marked "effective" resolution.

Assuming he is correct, then our TRV-950, in today's language, is a 2.0 MP effective camera.

Combined with a 12X optically stabilized zoom, powerful image controls, and multiple focus methods, a very good digital camera in good light.

[Kenn Jolemore]

One thing is for sure I have shot some wonderfull pictures with the 950 and had no problems printing them at 8.5x11 high gloss with tremendous results.Matted they do well at sidewalk fairs and such. If anyone would like to see a sample of some Yellow Night Herons shots I took(for example) feel free to e-mail me for a pic.
KennJ