Specifics

We now take up the question of 24 bit color versus 48 bit color.

There are two modes of color evaluation and reproduction. For images that are intended to be transmitted, the mode breaks color images into their red, green and blue components. For reflected images, the mode breaks color images into cyan, yellow, magenta, and adds direct luminence control with the color black. The former mode is known as RGB, and the latter, CYMK. In this matter, it is RGB mode that is used.

Color, then, is evaluated and recorded, in this case digitally, with the use of three channels. Of specific interest in this investigation is the question of the information capacity of these channels. For each picture element (pixel), a discrete chunk of information is generated. In computer science, the technical term for this information chunk is the 'word', and is quatified by the number of binary digits (bits) it comprises. In short, then, it is the number of bits that are at issue.

The traditional length of a 'word' is eight (8) bits; this is also known as a 'byte'. But here, the concept of the digital word is used to describe the size of the physical registers that a computer uses to store information. In general, though not always, the bit length of a computer register has been a multiple of two. When computer software directly interfaced with the hardware, the primary determinate of register length was the amount of information needed to route the register throughout the hardware, including the address and state of various busses and ports. Now, customarily, it is only the operating system itself that interacts with the hardware, and so word size has been largely standardized as multiples of the byte; the determinate here being primarily the requirements of the computer architecture itself, including the addressing of information storage. Hence, the standardization of a word as the byte, as defined in the C programming language (the traditional language of system programming).

Thus it is that the default word length for much software now available, and the register size of many peripherals, is the byte, or 8 bits. When the word size is increased, it is increased in multiples of bytes, and so it is that the first increase of word size is from one to two bytes, or from 8 bits to 16 bits.

So, the default color mode comprises three channels of 8 bits each, or 24 bits in total. The alternative most commonly available comprises three channels of 16 bits each, or 48 bits in total. The question is: what is the significant difference between these two modes? The answer is: the amount of information each will carry.

An 8 bit word can carry 256 different values (2^8, or two raised to the eighth power). A 16 bit word can carry 65,536 different values (2^16). This is a very significant difference in information carrying capacity. When we sum the capacity of the three channels, the difference increases dramatically. Three channels of 8 bit capacity can represent 16777216 different values (256^3). Three channels of 16 bit capacity can represent approximately 2.8 * 10^14 different values (65536^3).

With regard the matter at hand, these different values represent increments in a gradation of luminosity of specific wavelengths of light, as recorded by a light sensor. Incidentally, not in issue here is the wavelength itself, but rather the sensitivity of the sensor, specifically its ability to discriminate differences in luminosity. Central to this investigation is the apparent fact that no available, or at least handy, means exist wherewith to quantify this sensitivity. We must assume that different sensors have different capabilities for several reasons, and that different scanners use different sensor arrays with differences in capability for different sorts of reasons. So we get to the fundamental question here: can we expect sensors to record more than 256 different levels of luminosity?

If we can, then the use of 8 bit channels becomes a limiting factor. If not, then 8 bit channels are satisfactory. Note here that no thought is given to the notion that 16 bits of information will be fully utilized, only that more than 8 bits are needed to fully satisfy the recording capabilities of a given sensor.

Again, we are not interested in the technical specifications of the devices in use here, only their observed performance. So we can ignore such information. What we are interested in here is what may be the sources of such differences as are observed, and so we turn our attention to what those sources might be.