Home  |  News  |  Reviews  | About Search :  HardWare.fr 

MiscellaneousStorageGraphics CardsMotherboardsProcessors
Advertise on BeHardware.com
Review index:
Test: the influence of electronic components on LCD colors
by Vincent Alzieu
Published on January 31, 2006

8, 10, 14 bit dithering?
Four images were examined: an 8 bit pattern, an 8 bit with dithering (color scale + progressive), a 16 million color pattern (8 bits, as well) and a 12 bit medical image.

Amongst the four monitors only one claims to have precise colors with the Eizo L887´s characteristics mentioning an internal 14 bit processing. We haven´t really successfully obtained details on this subject, because, in fact, the monitor only has an 8 bit input. After that there is a 14 bit interpolation that is probably more intelligent than just adding 6 zeros behind 8 bit values. For this the monitor has to process the image dithering, recalculate it and base this calculation on surrounding pixels to establish more accurate averages.

A reminder on this subject might be useful. TN monitors are all 6 bit and FRC is sometimes listed in their specifications. FRC means dithering.

TN monitors process a 6 -> 8 bits dithering
6 bits = 64 levels per RVB component, 64 voltage values applied to liquid crystals. When games will use 16 million colors, using 64 x 64 x 64 = 262 K is inconceivable. A processor is in charge of interpolating the signal to 8 bits. It adds two zeros on the end each time:

Let´s take the case of an 8 bit game scene that requires displaying 250 colors. A 6 bit monitor without dithering would only successively display 248 and 252 colored pixels:

This flickering operates at 60 Hz, and so the eye would see the 252 pixels 30 times and the 248 pixels 30 times in one second. This would be so fast that we would see a mix of the two values at 250. At this frequency, there is the risk, however, just like with CRTs where there would be a visual disturbance related to this "twinkling".

In the end it would be better to work with four adjacent pixels and equivalently distribute the 248 and 252 pixels. Pixels are so small that the eye only sees one color, 250, in only one interval of time (1 Hz) without twinkling.

On the left, the eye sees (252 + 248 + 252 + 248)/4 = 250,
one the right : (252 + 252 + 252 + 248)/4 = 249

One important point here is that a 6 bit monitor that uses 8 bit dithering won´t work with 256 interpolated colors. The maximum value in 6 bits is 11 1111. If it is extended to 8 bit it gives 1111 1100 = 252.
253 red shades x 256 B x 253 G = 16.2 million colors and not 16.7. This is the reason for the differences between TN type monitors (all in 6 bits) and FCR, VA and IPS panels, which function in 8 bits.

Some IPS and VAs use 8 -> 10 bit dithering
The principle is exactly the same. With some monitors, manufacturers, buyers and graphic artists estimate that 256 shades aren´t enough and they aren´t necessarily wrong. We can easily admit that the eye´s limit is "only" 16 million colors. Let´s just say that this is quite convenient, because monitors are often incapable of displaying more. You have to carefully look at a pattern of 256 shades of gray to realise that the eye singles out all colors and doesn´t mix them. Here we see the colors and separations between the different shades:

Those who have good vision and a precise and accurately toned monitor can see vertical stripes of approximately 14 pixels. Once your eye gets used to this image you will see stripes on the whole length of the pattern.

If it is applied to a normal picture, this drawback can become rather annoying and unattractive, when we want the nicest images possible. This is when solarization defects appear, and instead of color scales there are areas of color.

This example created without dithering, jumps from one color area to another. We see a series of circles visible on all monitors whether they are in 6, 8 or 10 bits. To reduce these effects, it´s possible to introduce a diffusion of the error and mix the pixels of close colors in transition areas.

Here is another pattern, also with dithering used for tests (in a bigger size):

The above image deserves a couple of explanations:
- first it only includes shades of gray. Those who are unfortunate can see green and red circles. This was our own experience.
- unlike what some monitors display, this pattern represents a perfectly progressive color scale. Monitors that display a dark ball of a single color in the middle, have an incorrect image. This was the case of the NEC monitor when its brightness deviated from its initial value and had a direct impact on the accuracy of gray shades. Those who see permanent jumps of colors can try to better adjust their monitor by playing with the OSD gamma and/or contrast.

To compensate for this defect, Nec, Eizo, LaCie and Iiyama started to release 10 bit monitors. Actually, they are 8 bit + FRC:

Just like the above explanation on 6 ->8 bit dithering, there is a processor on these 10 bit monitors to diffuse errors and obtain colors not possible in standard 8 bit:

These 10 bits monitor (8 bits + FRC) display values from 00 0000 0000 to 11 1111 1100, or 0 to 1021 = 1022 nuances (and not 1024). 1022 x 1022 x 1022 = 1.07 billion colors. This is the reason for the theoretically better and more subtle rendering. It has nothing to do with panel technology, except that it only concerns VA type monitors (PVA, MVA, ASV) and IPS for now. In practice, the Eizo monitor generally displays more accurate and subtler color patterns than the others. We say, "generally", because it tends to change with time. At their best calibration, the three competitor monitors provide equivalent results. Sometimes they may stray, which leads us to believe that Eizo´s accuracy is more due to the stability of their brightness system than to its 14 bit dithering.

Dithering in practice
We displayed several color patterns on our four monitors in addition to10 bit black and white medical images.

8 bit grey patternFor some of the patterns all three calibrated monitors provide the same result. For others like the radial color scale shown above differences were radical. We easily established a quality order from this pattern:

n°1 Eizo L887, ASV Sharp panel, 8 bits native + FRC = 14 bits : radial color scale is perfect. Nothing to say here.

N°2 Belinea 10 20 35W, P-MVA AU-Optronics panel, 8 bits: the scale is good.
Sometimes when brightness varies we see a slight halo or soft but present demarcations for some color changes. It is nevertheless a good result, and you really have to look hard to see a difference with the Eizo monitor.

N°3 Hyundaï Q90U; TN panel Samsung, 6 bits + FRC = 8 bits: the result is very close to the Belinea´s with slightly more noticeable demarcations. Once again, they tend to vary with time. They change from perceptible to invisible when the monitor goes back to correct colors. At worst we clearly see a circle 2.5 cm away from the center of the image.

N°4 Nec´s LCD 2170NX S-PVA Samsung, 8 bits panel. The image varies from the best (perfect color scale) to the worst. There may be a 2.5 cm wide dark ball in the centre of the image that is strongly dissociated from the rest. Then it returns several hours later to the best.

12 bits medical images

The software provided by Fenics was developed for 8 bit monitors. For now there are only 8 bit panels on the market except for a couple of black and white products destined for the medical world. It was used to display 12 images, in TIFF format, each 34 MB.

To avoid defects noticed during previous test patterns, Fenics doesn´t just send the image as such. They bought a Mil Lite layer from Matrox, who is in charge of diffusing the error to have the best image as possible in 8 bits. In short, it processes a reversed dithering from 12 to 8 bits. In terms of quality, the result is surprising:

- images are almost perfect for all monitors.
- all monitors provides similar images (at least the four tested here). Eizo doesn´t display nicer images than the others this time. The rendering is the same for all.
- whether if it is with Eizo or Nec, we noticed slight color breaks only visible after a careful study of the panels. This border between two colors is probably due to the loss of information following the change from 12 to 8 bits though Matrox´s API.
- if the image with Matrox´s error diffusion is opened on two monitors, one in 8 bits and the other in 12 bits, the 12 bit image is the one most appreciated by users. They can´t explain their choice most of the time. Even if we carefully look at the images, it´s impossible to notice a real difference.

<< Previous page
H + 0 to H + 30 hours color quality

Page index
1 | 2 | 3 | 4 | 5
Next page >>

Copyright © 1997- Hardware.fr SARL. All rights reserved.
Read our privacy guidelines.