Video to photo print
 

Lets suppose I shoot video or digital cinema, but want to enlarge and print one video frame onto photo paper approximately 8x11 inches. What camera/format/codec would yield the best image quality on photo paper?

If you want to just grab a frame from the video, you're going to have a problem. An NTSC video frame's pixel size is fixed at 720x480 pixels by the NTSC standard and that's independent of the camera, format, codec, etc. If you expand that to fill an 8x10 or 8x11 photo print, your resulting resolution will only be ~65-72 dpi (dots per inch). This might be acceptable resolution (barely)for web viewing on a computer monitor or on video screen where motion blurs the perception of the pixel boundaries but it will be totally unacceptable on a hard copy print - the lack of sharpness and the pixelation will be very apparent. While the required resolution for a print depends both on its intended use and how picky you are, generally somewhere in the vicinity of 175-250dpi would be considered the rock-bottom starting point for simple amateur snapshots, while serious and commercial work should at least double or triple that.

If you have a camera that is still photo capable, use it to grab a few stills while you're still setup for the scene, or do what the studios do and have a still photographer on-set handling that side of things if you really need good quality photos, say for publicity shots etc.

Thanks for your reply, Steve.
I'm currently using a Sony Z1, hdv, and I've managed to grab video frames at 1440x1080. In photoshop, I de-interlace the image and bump it up to 300dpi. The resulting image blown up on 8x11 paper is actually not too bad. I would imagine, however, that a progressive frame capable camera would yield a better result.

But what about Silicon Imaging S12K? Or Red? How would a frame grab from these look?

The average human eye can resolve about 300 dpi, +/- 60 dpi. You can determine print sizes bydividing the resoltion by dpi (240 to 360 dpi). A 6 MP camera hAs resolution of 3000 x 2000. Divide 3000 by 300 and you get 10. 2000 divided by 300 is a little over 8. So, a 6 MP camera can easily produce an 8x10. Software interpolation can increase picture size with exceptable results.

Make sure that you also resize the image or the proportions will be wrong because HDV is anamorphic. The final image should have a width which is 1.78 times the height, such as 1920x1080 or 1440x810.

You are absolutely correct. But even a full HD frame isn't 6 megapixels, it's only a shade over 2 mp. As for "acceptable results" with software interpolation, that's a matter of hugely subjective interpretation. Acceptable for what purpose? Newsprint reproduction, press kits, production record prints, art gallery exhibition?

Also, a note of caution re megapixel ratings for consumer cameras. These cameras generally use 4 sensor pixels per image pixel ie 2 X Green 1 X red, 1 X Blue if they use the Beyer pattern

Large format digital backs actually use piezoelectric motors to shift the sensors during exposure - I think sometimes shifting in a 4 X 4 quarter-pixel pattern to get 16 images which are then merged into the final image.

Very nice resolution.Very high ($30k) price tags - just for the back. You need to already have the camera and lenses. Maybe $40k or $50k or more by the time you're done.

,

To determine resolution multiply vertical resolution by horizontal resolution. In my example, 3000 by 2000 equals 6 million. Agreed acceptable results is very subjective. But the degree of acceptance depends largely on two factors, ability of the eye to resolve the image and the distance from viewer to the print. An image from a newspaper can look sharper than the image from a $50K camera if the newsprint image is viewed from an optimum distance and the other image is viewed from a less than optimum distance.

Software can never increase resolution in an absolute manner. Software works in ways that trick the eye into perceiving that the image has more resolution than it actually has. The human eye responds to parts of the image in different ways. The eye perceives more detail in the luminance portion of the picture than the chrominance. Course detail is better than fine detail. The JPEG algorithm and the Bayer (Dr. Bryce Bayer) pattern used in digital cameras make use of both precepts. Software tricks generally allow the eye to receive detail that is not in the image, in an absolute sense. A 6 MP camera can generally be interpolated to 11x14 with no loss of quality to the human eye (when viewed from an optimal distance). A 2 MP camera (1600x1200) will produce 5x7 images.

I seem to recall that Bruce Bayer was at Kodak when he developed the Bayer patern. And just recently Kodak announced a new filter technilogy that may have the effect of increasing the light sensitivity of the sensor array significantly.

Also, you raise another interesting point - the distinction between resolution and perceived sharpness. Quite different animals indeed.

No matter how you slice it, a still image from a nrmal video camera isn't likely to win any prizes for appearance, but then again the video camera is designed for a completely different purpose than a still camera.

I find it interesting that most higher end 3D animation packages will introduce intentional blurriness into the otherwise crisp CG images in order to get smoother appearing animations.

By the way, the annual computer graphics big deal event (SIGGRAPH) is in San Diego next week and well worth a visit by anyone in the area. 30 or 40 years ago it was a group pf PhDs arguing about how to draw a line on a computer monitor, and now it's one of the biggest shows around attended by all the major movie studios as well as all the CG software and hardware crowd. Fascinating and highly recommended for anyone in the area.