• Blogs
• Tradeshows
• Partners
• White Papers
• Events

This article contains tables only available in PDF format. To view it, you must have the Adobe Acrobat Reader, which can be downloaded for free.

With new CCD technologies arriving, the issue of single-CCD vs. three-CCD cameras has been pushed into the limelight.

The introduction of the JVC JY-HD10 has renewed ongoing questions about image quality and single-CCD cameras.

JVC's recent announcement of the GR-HD1 and JY-HD10 low-cost, high-definition camcorders has brought forth an avalanche of controversy about the technology required for the capture and recording of high-quality images. At first, questions centered on what was perceived as a far too low MPEG-2 recording data rate. More recently, the issue of the number of CCD chips has come to the fore. This issue, in fact, has been lurking in the background for the last year.

Many have been shooting photographs with single-CCD cameras. I have found that the pictures have excellent color quality, even though the cameras use only a single CCD.

Also, many have seen the announcement by Foveon about its X3 technology, which allows 35mm film-quality images to be captured by a single chip. The Foveon image sensor employs three layers of photo detectors embedded in silicon. The layers take advantage of the fact that silicon absorbs different wavelengths of light at different depths, so the top layer records blue, the middle layer records green, and the bottom layer records red. This means that for every pixel location on Foveon, there is a stack of three photo detectors.

Three-chip cameras also capture all three colors from each spatial location. Single-CCD cameras do not. Single-chip cameras employ one of three filter schemes: Bayer (primary color), Com-plementary color, or JVC's new Hybrid Complementary-Primary design. All make use of a 2×2 CCD element cluster to obtain luminance and chrominance information as shown in Chart 1.

Because of the interest in the JVC JY-HD10 camcorder, I'll use it as an example of a contemporary, single-CCD camera.

The new JVC design uses clear, green, cyan, and yellow filters to deliver maximum vertical resolution as appropriate for a progressive scan (480p and 720p) camera. The filter matrix employs two complementary color filters (Ye and Cy) with one primary (Gr) color filter. Chart 2 shows how adding any two complementary colors (Mg, Ye, and Cy) yields white.

Chart 3 shows how luma (Ln) samples and chroma samples (RGBn) are generated by the JVC's CCD elements (En). Luma is obtained from the color mosaic using the following algebra: Y = .29R + .59G + .11B or Y = R + 2G + B. Odd row luma samples are calculated as Yodd = W + G. White, of course, equals R + G + B. Thus, the White filter elements plus the Green filter elements yield Yodd = R + 2G + B. Even row luma samples are calculated as Yeven = Cy + Ye, where Cy = G + B and Ye = R + G. Thus the Cyan filter elements plus the Yellow filter elements yield Yeven = R + 2G + B.

Chart 3 shows how 12 luma samples (H-1 × V) are generated from the 16 (H × V) elements. Using this formula, the JVC chip in 720p mode generates 842,861 luma samples from 659 CCD rows. (The camera's DSP scales the image up to 720 lines.)

JVC CCD Operation

The chroma generation process requires the upper CCD row of each pair of CCD rows to be delayed one video line. The delayed line is then available to be combined with the current line. Red is obtained via R = W - (G + B) where (G + B) = Cy. Using all four filters for greater sensitivity and equal matrix sampling, R = (W + Ye) - (G + Cy). Blue is obtained via B = W - (R + G) where (R + G) = Ye. Again using all four filters, B = (W + Cy) - (G + Ye). Green is obtained via G = W - (B + R) where (B + R) = Mg. There is no Mg filter, but we can use B and R samples. And once again using all four filters for greater sensitivity and equal matrix sampling, G = (G + Cy + Ye) - W.

Note in Chart 3 that nine chroma samples (H-1 × V-1) are generated from the 16 (H × V) elements. Using this formula, the chip in 720p mode generates 841,582 RGB chroma samples.

Looked at purely from the number of luma and chroma samples generated, as Chart 4 shows, there is a very small difference between single- and three-chip 720p cameras.

To understand the real difference between camera designs, one must look beyond the number of samples generated from the CCD elements. Rather, you need to look at the spatial nature of elements that are combined to create a sample. Doing so will provide a count of the number of pixels generated, which determines the camera's horizontal and vertical resolution.

Let's start by looking at a conventional NTSC CCD that captures 481 rows of information during a time period determined by the camera's shutter-speed setting. The CCD is then read out from the top row to the bottom row during 1/60^th of a second. One interlace field is obtained from the upper 480 rows. To capture a frame's field-mate, once again the CCD captures the image during the shutter exposure time, and then over the next 1/60^th of a second, the lower 480 rows are read out from the CCD.

Each element from a row is delayed one “line time” and added to the equivalent element in the next lower line. By adding data from two successive rows together, a filter is created that softens horizontal edges, thereby reducing interline flicker. When signals from two CCD elements are summed, signal strength is increased. And appropriately, the filtering process outputs 240 video lines — the number of lines required for an NTSC field.

A price must be paid for the benefits of a sliding-row filter. The filter decreases image vertical resolution by about 25% to 180 lines per field. Thus, the effective vertical resolution of an interlaced NTSC frame is reduced to about 360 lines. And in DV mode, the JVC camcorder is measured at 360 vertical lines of resolution.

It's important to note that when an interlace scanned image is recorded, the sliding filter is used no matter whether the camera uses one or three CCDs. Conversely, when a progressive image is recorded — no matter the number of chips employed — a sliding-row filter is not used to obtain luminance information. (However, the filter can be activated; as when a Panasonic AG-DVX100 is switched to “thick” mode.) When the JVC camera is in 480p mode, its vertical resolution is measured at 480 vertical lines.

All single-CCD cameras generate luma samples by means of a sliding filter that moves across pairs of CCD columns. This filter decreases horizontal resolution by about 25%. Therefore, 720 elements (DV mode) yield a video row with 540 pixels; 941 elements (SD mode) yield a video row with 706 pixels; and 1,280 elements (HD mode) yield a video row with 960 pixels. Unfortunately, a second process is at work that causes measured resolution to be less than expected from these pixel counts.

CCD-based cameras utilize optical and electrical low-pass filters to reduce aliasing that naturally occurs from the discrete sampling of an image by the array of CCD elements. The filter's cut-off frequency is a function of the number of CCD horizontal elements. Note that vertical resolution is not affected by the anti-aliasing filter because the cut-off frequency is set based on the far higher number of horizontal CCD elements.

The mosaic array in front of a single CCD increases the severity of aliasing and so requires that the cut-off frequency be lowered, resulting in a loss of measured horizontal luminance resolution. The JVC camera has an anti-aliasing filter that reduces fine detail — and only fine detail — by about 25% for both 480p and 720p scanned images. (The cut-off frequency for the anti-aliasing filter is too high to reduce detail for the relatively lower-resolution DV image.)

Chart 5 shows vertical and horizontal resolution measures for the JVC cam-corder. The measured SD and HD horizontal resolution values reflect two consecutive 25% reductions in fine detail.

Ignoring the effects of the anti-aliasing filter, how much greater HD resolution would be available if three CCDs were used rather a single CCD? There are two ways to answer this. At the superficial level, because a sliding filter would not be used, the potential horizontal resolution would increase, in 720p mode, from 960 pixels to 1,280 pixels. However, that is a measure of luminance performance. Most concerns about single-CCD cameras relate to their expected poor chroma performance.

Let's look at a simple way to compare three-chip and single-chip cameras. What's the ratio of chrominance to luminance pixels for a three-CCD camera? The chroma/luma ratio is 1:1, or 1.00.

The JVC chip provides 632,640 luma pixels (960×659 pixels). Each RGB sample is obtained from information from two CCD columns using a sliding two-sample window. Thus, horizontal chroma resolution is decreased by 25% from 1,280 elements to about 960 pixels. A sliding two-sample window is also used in the vertical direction. Thus, vertical chroma resolution is decreased by 25% from 659 samples to about 494 RGB pixels. Total RGB chroma “resolution” is, therefore, 474,240 pixels.

The ratio of chroma pixels to luma pixels is 474,240/632,640 = .75. Thus, the JVC chip's chroma/luma ratio is lower than that of a three-CCD camcorder.

Chroma detail, however, must be further reduced before it is recorded as color difference components. The reduction required is the same for DV (4:1:1) and MPEG-2 (4:2:0). And this reduction is required for both one- and three-CCD cameras.

Prior to recording, the RGB chroma information undergoes two processes. The Red, Green, and Blue samples are combined (.29R + .59G + .11B) to generate alternate luma samples, Y'. These luma samples are then mathematically combined with the Red and Blue samples to create two color difference components: R-Y' and B-Y'.

At the same time, for both DV and MPEG-2 compression, the chrominance detail is reduced by one-quarter. Therefore, with DV's 4:1:1 sampling structure, R-Y' and B-Y' color samples for every fourth luminance pixel on every line are recorded. MPEG-2 Main Profile, with its 4:2:0 sampling structure, records R-Y' and B-Y' color samples for every other luminance pixel on every other line. Both sampling spaces represent a chroma resolution reduction by a factor of four — attenuating the chroma resolution advantage of a three-CCD camera.

So, are three imaging chips necessary for high-quality video? Cameras with three CCDs do offer greater color quality — plus eliminate color artifacts that can occur with rapidly moving objects. However, with enough resolution, a single CCD will support oversampling. Oversampling can eliminate the negative effects of filtering performed prior to compression or encoding. (Sony, for example, currently offers camcorders with a 2.11 megapixel CCD.)

Unfortunately, the drive to decrease CCD size and the need for multi-mega-pixel CCDs are in conflict, leading to a decrease in light sensitivity and light latitude. Lack of latitude increases the probability of lost shadow detail or highlight burn-out, something you don't want in an image. Lack of sensitivity can, under low illumination, increase image noise.

I believe technology advances may enable single-CCD video cameras to become the norm. And not simply for low-cost HD camcorders. And that's because there are limitations to cameras that employ three CCDs.

Three chips and an optical-prism add bulk and cost. The prism also limits the maximum F-stop of the lens. Moreover, where video is shot rather than film, it's difficult to shoot with a desirable shallow focus with current three-CCD cameras. A single 35mm chip, such as the 4,000×2,000 (8-megapixel) prototype image sensor from Dalsa (see page 80 in the March 2003 issue of Millimeter) would solve these problems.

Therefore, whether future CCD chips are large or small, technology advances will force us to stop counting CCDs in our efforts to prejudge camera quality.

---