Resolution Test Patterns

Sony TRV900 (3-chip MiniDV) and TR7000 (1-chip Digital 8)


I took some pictures of the EIA 1956 test pattern to measure resolution (for high-resolution cameras, you should probably use these better test patterns.) The images below are details from the full frame, which you can see by clicking on it.

Sony TRV900
Sony TR7000

These test patterns were shot indoors with a 1 kW halogen flood for light. Both cameras were tripod-mounted, 5 feet from the target. Autofocus and autoexposure was used in both cases (I tried manual focus but auto worked just as well on both cameras). TRV900: f/5.6, 1/60 sec. TR7000: f/2.4, 1/100 sec, steadyshot on (steadyshot off: looks same but at f/3.4, 1/60). No filters on either camera. The TRV900 was manually white-balanced (the TR7000 is auto-white balance only). Both are shot as normal, interlaced scan video to tape, then played back into a firewire capture card on my PC and captured as a full frame. Because the TR7000 doesn't do progressive scan, you would need a capture board (firewire or analog) to capture full still frame from the TR7000; using the "photo" button while recording, or the "pause" button on playback, you are getting real image data only from a single field, and every other scan line in the video output is interpolated (making diagonal lines have visible steps or jaggies).

This image is from tape playback (not "live"), through the DV (firewire) port into the computer in each case. I did expand the histogram to fill the range in each case, but did not change gamma or apply any sharpen or other filter. Photoshop saved JPEG as Quality=8. This is the original 720x480 DV pixel format so the aspect ratio will be wrong on your computer monitor (slightly stretched horizontally) but this is the only way to show you the full original data.

At first glance I'd say the TRV900 does about 550 lines and the TR7000 about 350, maybe. However I know (now) the theoretical max. for DV at 720x480 pixels, is 720 * 0.75 = 540 lines (the 0.75 factor is because video resolution is traditionally defined "per picture height" and the image has a 4:3 aspect ratio). Apparent resolution beyond 540 lines is an artifact of a finite number of pixels, and having only 4 converging lines on the chart; if you had more lines in the pattern, you'd notice that you start loosing visible lines beyond the 540 mark. Anyway examine the images and judge for yourself. If you download the image, open it in a paint program and zoom way in on the converging lines, you will notice that the TRV900 image maintains resolution right up to the level where the pixel row is black-white-black-white, that is, it's just about as sharp as the standard 720x480 DV signal can ever get, however many "lines" you consider that to be. By the way, the TRV900 can actually do better than this when used "live"; see this note.

It happens that the TRV900 shot shown is "best case" because it is aligned for the best resolution. If you shift the camera 0.5 pixels to the left, the black-white-black-white line becomes grey-grey-grey etc. since each CCD pixel sees half-white and half-black. The resulting apparent resolution change is known as the "Kell factor". It turns out the conventional definition of resolution (540 lines max for DV, as compared with 400 for SVHS, 240 for VHS, etc.) is a "best case" number, and does not take the Kell factor into account. For the analog formats it's irrelevant since those aren't pixellized horizontally, anyway. Resolution turns out to be more complicated that you might think. For a more detailed discussion of resolution, see Allan Jayne's Video Resolution and also Peter Utz's Understanding Resolution.

By the way, running the S-Video out from the TRV900 to my 27" JVC AV-27920 television (advertised at 450 lines resolution) shows only 250 or 300 lines at best. Hmm... looks like I want a better TV.

This pattern is the most standard resolution test pattern used for testing video cameras, the "EIA RESOLUTION CHART 1956". This specimen is white film, 11.5 x 8.5 inches, mounted on cardboard. It is 5 feet from the camera and lit by a 1 kW halogen flood fixture about 5 feet behind the camera. I fixed the camera and adjusted the zoom for the white triangles along the edges to just touch the border of the (viewfinder) frame. Ideally all borders would touch if the viewfinder had a 4:3 aspect ratio. The TRV900 viewscreen is slightly wider, and the TR7000 viewfinder is slightly taller, than perfect 4:3.

I framed for the closest shot in which all edge triangles were visible. The computer images from DV show the cardboard mount area which is in the "overscan" region on your TV and not normally visible. A broadcast engineer told me that the proper method is to fill the entire image with the sheet, including the overscan region. Given you don't see that region in the viewfinder, I guess you need either a computer display, or a pro monitor with overscan turned off, to frame it exactly.


I have some earlier shots from a lower-resolution test pattern here. In those, the TR7000 color aliasing isn't as evident. They were also shot outdoors (diffuse, natural sunlight) which seems to make a big difference. I originally thought the TR7000 was more sensitive than the TRV900, but my tests today using this test pattern showed the TRV900 to be about one stop more sensitive in both natural and artificial light. It may be that the cameras' auto-exposure systems work differently on a blank white wall (my initial test), and on a scene with lots of contrast (test pattern).

The test pattern shown above is in real life a perfectly neutral greyscale, with no apparent color. You will note some yellow and purple colors on thin horizontal lines from the TR7000 not present with the TRV900, due to 1 CCD chip vs. 3 chips. This shows up most clearly on high contrast, small features (such as this test pattern), particularly when lit by artificial light. In natural lighting the problem is less evident, see this shot for a comparison of natural light, evening twilight in this case, and a halogen lamp. You'll also note that the auto-white balance doesn't remove the blue tone of the evening light. The TR7000 is about 1/3 the price of the TRV900 and you can probably see why.

Gerhard Pfeiffer posted this TRV110e test captured via S-Video (analog) using a Hauppauge WinTV capture card (he's also got some nice flower pictures). My manual says the (NTSC) TR7000 uses 290k active CCD pixels, and I'm told the PAL version has 400k active pixels, which is a significant difference.


Do your own test

  For high-definition cameras or DSLRs, I suggest these test patterns  For "standard definition" cameras, if you want to print out your own test pattern, you can download this scan of the EIA 1956 standard test pattern as a bitmap: 1440x1086 pixels or better yet, as a vector format (Corel, Illustrator, and EPS) EIA1956-v3.zip or in PDF format EIA1956.pdf. Thanks to George Ou for an extremely accurate vector conversion! Note, you can download a PS/EPS/PDF viewer from the Ghostscript website, or use the Adobe Acrobat Reader for PDF files. (The resolution test PDF file may look poor in your browser window, but it will be clean on a high-resolution printer.)

Print out the pattern at a resolution that fills the page (without cropping off anything), tape to a wall, and frame it with your camera so the edge triangles just touch the borders of your image. If you do this, the point where the converging lines in the resolution wedges cannot be distinguished anymore, is the number of "lines of resolution" of your camera. (This would be the 0% modulation or limiting resolution point, where lines and spaces both become 50% grey. The more strict professional definition of resolution specifies the 50% modulation point, where instead of full white and full black, you have 75% and 25% video levels on the lines and spaces. Normally you use a waveform monitor to measure these video levels, although with digital video you could just compare pixel values from a still frame using a paint program.)

Video resolution is defined as number of (light+dark) lines resolved horizontally, divided by the image ratio which is here 4:3. The MiniDV format can, in theory, resolve 720 pixels / (4/3) = 540 lines. This resolution definition is independent of actual image scale, so you can do this test with any size printout at any distance/zoom setting and get comparable numbers, just as long as the pattern exactly fills the full frame. Remember that almost all camcorders, including the TRV900, do not quite show you the full image area on either viewfinder or viewscreen: there is a hidden "overscan" region of about 10% on each edge. The image captured via memory device or firewire does show the full area.

Where this file came from: I scanned the original at 600 dpi on my Microtek E6 flatbed. That gave me an image about 7000 by 5000 pixels (35 Meg). I fixed some dust spots, cut the resolution down, and converted to JPEG in Photoshop. I increased the contrast for the best looking thin lines in a JPEG image, so the greyscale blocks on the JPEG files is all wrong. However, the vector format files do have a useful greyscale progression. The lines aren't perfect but this version should be useful for testing most video cameras. The vector formats in the zip file were hand-matched to the bitmap scan, and a printout of that version is very close to the original film specimen (if your printer is good enough!). If you want the real thing, you can obtain it from Edmund Scientific (Industrial Optics Division): EIA resolution chart, 8.5x11.5", black on white film (0.2mm thick), part H52997, $80.


Postscript pattern, also in PDF format respat.pdf. Here is a different resolution test pattern written by Dan Tomandl in Postscript, which comes with notes in the file on how to use it. It can be printed out at any resolution you like (if you have a postscript printer) or use a viewer like Ghostview. This pattern is useful for testing printers, film recorders, and other devices as well- the calibration marks go to 4800 lines, but the wheel pattern goes down to a mathematical point at the center so there's no practical limit to the resolution. It's interesting, for good-looking lines in greyscale you need to oversample by a factor of 8 or more. That is, to make a clean 360 dpi image (3960x3060 pixels for 8.5x11" image) I had to render it (in B/W) at 2880 dpi (0.78 billion pixels total!) and downscale by 8x. The 360 dpi greyscale image is pretty clean up to about 1200 lines of resolution (well over the 750 line limit of a broadcast TV camera). I'd post the result here but even the compressed .PNG image is over 1 MB in size. Have fun generating your own!
Wedge pattern as FrameMaker4 format reswedge.doc, also in EPS format, as a PDF document or a lower resolution bitmap in PNG format. This is intended for a 8.5x11" US letter size page, and prints resolution bars using nearly the full page, useful in case you want to test resolution from a long distance away. It is calibrated in horizontal line pairs per (arbitrary unit) so it is useful for relative (not absolute) resolution measurements.

The ISO-12233 pattern is a more comprehensive test chart than EIA1956, it is adequate for HDTV and other high-res devices. Software such as Imatest can automatically measure camera resolution using a slant-edge image. Displaymate.com offers a collection of more test patterns. You can download the USAF 1951 test target here. Here is a color chart with 24 standard colors.


comparison images of cats.
Back to TRV900 page.
Back to Digital 8 page.