High Bit Images

The High Bit Image

1) High Bits: What they are 2) Working with high bit colour 3) Posterization

Many people have asked me why I recommend scanning and working in high bit mode (12 to 16 bits per channel) when they can not see a difference between an image at 16 bits and the same file at 8 bits. Both look exactly the same on their monitors, and the 16 bit image will not print until they convert to 8 bits anyway. The question is fair, straightforward, and simple. The answer is rather involved and begins with the need for me to ask you a couple questions.

Can your scanner not output more than 8 bits per channel to your image editor? This can be called 10 bits per channel, 12, 16, or simply High Bit. Even if the maker advertised it as a 36 bit scanner, it may not output more than 24 (8 bits per channel times 3 channels). This is a fairly common "selling point", like advertising the highest possible dpi setting, even though it is useless and meaningless. If your scanner will only output 24 bits - you don't have to know anything about this stuff - yet.

Can your image editor work with more than 8 bits per channel? If not, class dismissed. I know Photoshop, Photo-Paint, and Paint Shop Pro, can work with 16 bits per channel. If you are using something else you'll have to find out on your own.
If you answered Yes to both questions, or are merely curious, read on.

High Bits: What They Are and Why You Can't See Them

Colour on the computer is not analog. Much like digital discs which turn sound into numbers, the colour system in your computer is made up of discreet shades instead of the continuous rainbow of the real world. There are three colour guns in the back of the monitor, Red, Green and Blue. Every colour you see on the screen is made up of a combination of the three base colours from these guns. The output of these guns is controlled by the graphics card, which can send 256 discreet signal levels to each gun. This is 8 bits per channel.

How does eight equal 256? As with every other aspect of computers, binary math is the answer. The largest number that can be described in eight places - 11111111 is binary for 255. Since zero is also a number that makes a total of 256. So you have a gun producing 256 shades of red, whilst another produces 256 shades of green and the third does the same number of blues. They are tossing out little dots of colour to the pixels on the screen. If you look at the colour picker in your program you will find that white is the following combination:

Red 255 Blue 255 Green 255

All three guns firing at their highest level. Black is the opposite of white at 0,0,0, Red is 255 RED, with 0 BLUE and 0 GREEN. These 256 levels of three colours is all your monitor can display. It cannot show more colours that this – unless you have a very expensive graphics workstation – in which case you should be writing this article not reading it.

Photoshop can do certain operations on images with 16 bits per channel. This may not sound like a lot more than eight, but it is. 4096 shades of each colour. Still doesn't sound like much more? Try doing a bit of multiplication and see how many colours you get with 256 times 256 times 256. Now do the same with 4096 times 4096 times 4096. Go ahead. I'll wait for you here.
Cheaters answer: 8 bit is called Millions of Colours. 16 bit is called Billions of Colours.

So what possible use are these Billions Of Colours? You can't see them, feel them, taste them, or hear them, but you know they are there. What can you do with them other than watch your hard disc struggle to save your bit bloated picture? I knew you'd ask.

Working With High Bit Colour

Lets pre-suppose a picture taken by the dim light of a couple sixty watt bulbs. The entire slide is red - and only red. There is an image there, but no apparent colour information (other than red). To add to the general malaise there is a fair amount of camera shake in the picture, and the subject was moving. There have been floods and a fire, and bad storage also taking a toll on the picture. It was shot on old Ektachrome slide film, and the emulsion is beginning to fall apart.

This slide is a real "tosser". But suppose this is also the only surviving picture you have of your daughter at the age of eight. How much would you pay to get a print from this slide?

It will cost you exactly 16 bits.

Posterization: Enemy of the Picture

If you have only 8 bits per channel the number of colour corrections you can make to an image is limited. Every move made to correct the deficiencies of the image will also lose some of the shades in the original. There is no way around this, nothing can be done about it. For every gain there is a loss.

The human eye can detect very small differences in colour, and can see when the tones "jump" instead of making a smooth transition. This is called posterization since it turns a photograph into something that looks more like a poster. Most image editing software has a Poster setting. Photoshop's is under the Image menu. Open a photo, preferably one of medium contrast with a full range of tones. Save it to a new name as we are going to destroy it by posterization. Now, simply plug in a number from 1 to 255 and let Photoshop show you the results - don't hit OK, and make sure the preview is enabled. High numbers, over 150, probably will not change the on screen looks of the photo at all.

Low numbers like 20 will give you dramatic change. These numbers tell you how many of your original 256 shades Photoshop is keeping. It discards all other shades, leaving only the number you specify. Cancel the command and let's take a look at how this really works. Also under the Image menu is a command called Histogram. When you open it up you see a bar graph representing all the pixels in the image.

The graph opens in the Luminosity Channel, which shows exactly how many pixels are at each of the 256 levels in the combined Red, Green and Blue channels of the photo. There are also separate graphs for each of the channels. If your picture is an average scene the graph should look like a mountain range, with peaks and lows, and smooth transitions between them. Close the histogram. Go back to Posterize and plug in 50 levels - hit Okay.

Now go back to the Histogram. The Luminosity channel still shows a complete bar graph, but there are "spikes" in some places. This is because the Luminosity Channel is made up of all three colour channels. Try the Red Channel. Not many lines in that graph, are there? Of the approximately 256 original levels of red bars in the graph, only 50 are left, and they are about evenly spaced across the breadth of the graph. Blue and Green look the same. Undo to a point before you posterized or go back to your saved version, and give it a try with 128 levels.

The photo may not show any change at all, the Luminosity channel may show only marginal damage. The Colour Channels however, will show as many gaps as lines. Every gap is a potential problem. You've already seen what can happen if you have too many of these gaps.

By starting with 4096 shades of each colour instead of 256 we can make several changes to the image and still have a full histogram when the picture is converted to 8 bits per channel to print.

Using the curves and levels commands on the red slide produce this.

This much restoration took ten or more moves in Levels and Curves. Had I started with an 8 bit per channel image there would be little left.

The next article in the series is a description of the Digital Workflow - How to have your digital cake and eat it too.