A few scanning tips

www.scantips.com

Color Bit-Depth, & Memory Cost of Images

Large photo images consume large memory and can make our computers struggle. Memory cost for an image is computed from the image size.

For a 6x4 inch image printed at 300 dpi, the image size is calculated as:

  (6 inches × 300 dpi) × (4 inches × 300 dpi) = 1800 × 1200 pixels

1800 × 1200 pixels is 1800×1200 = 2,160,000 pixels (2 megapixels).

The memory cost for this RGB color image is:

  1800 × 1200 × 3 = 6.48 million bytes.

The last "× 3" is for 3 bytes of RGB color information per pixel for 24 bit color (3 RGB values per pixel, one 8-bit byte for each RGB value, which totals 24 bit color).

The compressed JPG file will be smaller (maybe 10% of that size), selected by our choice for JPG Quality, but the smaller it is, the worse the image quality. The larger it is, the better the image quality. If uncompressed, it is three bytes per pixel.

Different color modes have different size data values, as shown.

Image TypeBytes per pixelPossible colorsFile Types
1 bit Line art1/8 byte per pixel
(1 bit per pixel)
2 colors,
black or white
TIF, PNG
8 bit Indexed ColorUp to 1 byte per pixel256 colors maximum. Graphics use todayTIF, PNG, GIF
8 bit Grayscale1 byte per pixel256 shades of grayJPG, TIF, PNG
16 bit Grayscale2 bytes per pixel65636 shades of grayTIF, PNG
24 bit RGB3 bytes per pixel (one byte for each of R, G, B) 16.77 million colors.
The "Norm" for photos, e.g., JPG
JPG, TIF, PNG
32 bit CMYK4 bytes per pixel, for PrepressVaries, but low. One guess has been approx 160,000 colorsTIF
48 bit RGB6 bytes per pixel 2.81 trillion colorsTIF, PNG


File Type PropertyJPG   TIF   PNG   GIF  
Lossy compressionX
Lossless compression   XXX
Uncompressed option X
GrayscaleXXXX
RGB colorXXX
8 bit color (24 bit data)   XXX
16 bit color option XX
CMYK or LAB colorX
Indexed color option XXX
Transparency option XX
Animation option X

RGB (Red, Green, Blue)
(0,0,0) (128,128,128) Middle Gray (255,255,255)
(255,0,0)
Red
(0,255,0)
Green
(0,0,255)
Blue
(150,0,0)
Darker Red
(0,150,0)
Darker Green
(0,0,150)
Darker Blue
(0,255,255)
Cyan
(255,0,255)
Magenta
(255,255,0)
Yellow
(0,83,0)
83
(0,85,0)
85, +2.41%
(0,87,0)
87, +2.35%

Understanding RGB Color

These are the basic definitions of these topics, and digital basics has more in other areas like resolution.

JPG files are 24-bit color in a 8-bit color system. These are two definitions with two different meanings. RGB color has three channels, and a 8-bit system means each Red, Green, and Blue channel is 8-bits (24 bits total of three). The 8-bits means 256 values or shades, values 0..255 (28 = 256) of each of Red, Green, and Blue. This can create 256 x 256 x 256 = 16.77 million possible color combinations, called 24-bit color. Today, the vast extreme majority of our monitors, video cards and printers are 8 bit devices, meaning 24-bit color, meaning they can show 16.77 million colors.

This 16.77 million is a calculated number, our eye cannot differentiate all of them. It is said that tone values must be 1% different for our eye to be able to separate them ("Just Noticeable Difference"). That's 100 exponential steps, but RBG tones have 256 linear steps. For example in the green samples at 85, a step of 2 might be detected as different if you try hard enough. It is 2%, and you probably can convince yourself you can at least detect an edge between them. The color is dark, but that helps the percentage. If double at 170, the percentage of the same step is half. Color is wonderful, and offers much, but difference in adjacent numerical values can be very subtle. And our eyes and brain do strange things to us. I can imagine the middle gray background can appear a bit lighter near the Black box and a bit darker near the White box, yet the background is exactly the same everywhere. The gradient strips show the 256 tones, which is the idea of RGB colors.

RGB colors are combinations of the three R,G,B basic colors. For example, we know from grade school that red and green make yellow. Or White is all colors. Or like RGB(238, 187, 68) meaning Red is 238 (of 255), Green is 187 (of 255) and Blue is 68 (of 255). There are 16.77 million possible combinations. That does Not mean any one image uses most of them, or that our eye can uniquely distinguish all of them, but they are choices. There's a good color picker from Mozilla to play with to match the colors to the numbers. The Photoshop Color Picker is a similar tool.

Our human eye does not perceive all colors as equally the same brightness. Of the three, we see Green as our brightest color. Red is next, and Blue is least bright (to us). Their perceived brightness (luminosity) is judged to be: White 1.0, Green 0.59, Red 0.3, Blue 0.11 (the three RGB adding to White's 1.0 when all are combined). That's the NTSC television formula for converting RGB to grayscale.

Grayscale mode is one channel of color, 256 shades of gray between black and white.

RGB(180, 180, 190) has a blue cast

RGB(180, 180, 180) is neutral gray

RGB(187, 180, 180) has a pink cast

RGB color mode simulates neutral gray with three equal RGB components. (80,80,80) or (180,180,180) are neutral grays. Neutral means equal RGB, no color cast. These are the same color as grayscale 80 or 180. White Balance tools use a known neutral white or gray white balance card this way to correct white balance in our color images. Clicking it identifies the spot, then the computer corrects the image so that known spot actually is neutral (equal RGB).

Grayscale files (single channel) are 1/3 size of color of RGB files, and have no color cast. We can print this with only one black ink with no risk of color casts, but which is a struggle with shades of gray (dithering is a way of combining ink dots with blank paper "dots" to create an averaged appearance of gray). Or we can print this with CMYK inks that are combined into shades of gray. Ink jet color printers also have to dither their four colors of ink. The few best printers for grayscale photos offer a couple more lighter gray inks.

Indexed color: However, as is the way, there is also another definition of 8-bit color (vs. 24-bit color). Indexed color (GIF files, and an option in TIF and PNG) use an indexed color palette with a maximum of 256 colors. Each pixel value is an 8-bit index pointing to that indexed color in the color palette (maximum of 256 24-bit colors). Not so good for photo images, but it's a very small file for a graphics of few colors. This was in common use before our current 24 bit color hardware became universal. More on Indexed color. A note from history, we might still see mentions of "web safe colors". It wasn't about security, this standard was back in the day when our 8 bit monitors could only show the few indexed colors. "Web safe" was six shades of each R,G,B (216), plus 40 system colors the OS might use. These colors would be rendered correctly, others were just a near match. It's conceivable a few old 8-bit video cards could still be in use, but those users should be used to imperfect colors now. "Web-safe" is obsolete now, every RGB color is "safe" for 24 bit color systems today.

Image size is always dimensioned in pixels. Data size is dimensioned in bytes. Data size varies too much to have any meaning about image size. Saying, the size of our 24 megapixel image is 6000x4000 pixels. This tells us how we can use that image. The file size may be 13 MB, but that doesn't tell us anything about the image, only about storage space or internet speed. A "regular mode" (most common) 24-bit color photo image is 3 bytes of RGB data per pixel. That means a 24 megapixel camera takes images of data size 72 million bytes (the calculator below converts this to 68.7 MB, uncompressed data size).

However, data Compression techniques can make this data smaller while stored in the file. In some cases drastically smaller, and maybe 72 MB goes into a 6 to 15 MB file if JPG compression. We can't state any exact size numbers, because when creating the JPG file (in camera or in editor), we can select different JPG Quality settings. And this compressed file size varies with image content too. Images containing much fine detail everywhere (a tree full of small leaves) will be a little larger, and images with much blank content (sky or walls, etc.) will be noticeably smaller (better compressed). But JPG files are typically 1/4 to 1/12 of this image data size. Both larger and smaller are possible. And then when opened, the image comes back out of the file uncompressed, and original size, with the original number of bytes and pixels when open in memory. It's kinda like magic. 13 MB JPG / 68.7 MB would be 19% original size, and we'd expect fine quality. 7 GB JPG / 68.7 GB would be compression to 10% size, and we would not expect best quality. But it could be an image with blank areas like sky or walls that compresses exceptionally well, which that is Not an issue.

But there are downsides with JPG, which is lossy compression, and image quality can be lost (not recoverable). Selecting higher JPG Quality is better image quality but a larger file size. Lower JPG Quality is a smaller file, but lower image quality. Don't cut off your nose to spite your face. Large is Good regarding JPG, the large one is still small. File size may matter when the file is stored, but image quality is important when we look at the image. Lower JPG quality causes JPG artifacts (lossy compression) which means the pixels may not all still be the same original color (image quality suffers from visible artifacts). There are the same original number of bytes and pixels when opened, but the original image quality may not be retained if JPG compression was too great. Most other types of file compression (including PNG and GIF and TIF LZW) are lossless, never any issue, but while impressive, they are not as dramatically effective (both vary greatly, perhaps 70% size instead of 10% size).

Image Size and Data Size Calculator

Numbers Only.  A NaN result means an entry was "Not a Number"
Image Width: pixels Image Height: pixels
Data Type:
If Printed at: pixels per inch

(Note that 24-bit RGB data is 3 bytes per pixel, regardless of how small the JPG file might be. See more detail).


One MB is a Bit More Than One Million Bytes

The memory size of images is often shown in megabytes. You may notice a little discrepancy from the number you calculate with WxHx3 bytes. This is because (as regarding memory sizes) "megabytes" and "millions of bytes" are not quite the same units.

Memory sizes in terms like KB, MB, GB, and TB count in units of 1024 bytes for one K, whereas humans count thousands in units of 1000.

A million is 1000x1000 = 1,000,000, powers of 10, or 106. But binary units normally are used for memory sizes, powers of 2, where one kilobyte is 1024 bytes, and a one megabyte is 1024x1024 = 1,048,576 bytes, or 220. So a number like 10 million bytes is 10,000,000 / (1024x1024) = 9.54 megabytes. One binary megabyte holds nearly 5% more bytes than one million, so there are about 5% fewer megabytes.

Calculator to Convert Memory
Bytes, KB, MB, GB, TB

Type a value somewhere here, and click its Convert button, to convert to all the other values.

Bytes B
Kilobytes KB
Megabytes MB
Gigabytes GB
Terabytes TB

If you ever see 1e-7, it means to move the decimal point 7 places to the left, 1e-7 to 0.0000001
Seeing a NaN result would mean the input is Not A Number.

Each line in the calculator is 1024 times the line below it. Which is binary, and is how memory computes byte addresses. However humans normally use 1000 units for their stuff. Specifically, megapixels and the GB or TB disk drive we buy do correctly use 1000 units (until we format them, when Windows shows 1024). Memory chips (including SSD) necessarily use 1024. File sizes do not need 1024 units, however it is normal practice. Windows may show file size either way, depending on location (File Explorer normally shows KB, but DIR shows actual bytes).

We also see units of Mb as a rate of bandwidth. Small b is bits, as in Mb/second of bandwidth. Capital B is bytes of data, as in MB size. Eight bits per byte, so Mb = MB x 8.

About Megabytes

Binary 1024 units are necessarily used for memory chips, but computer operating systems also like to arbitrarily use it for file sizes. All else (megapixels, purchased disk size, etc) use normal 1000 units.

Specifications for megapixels in digital cameras, and disk drive size in gigabytes are both correctly advertised as multiples of decimal thousands... millions are 1000x1000. Or giga is 1000x1000x1000. That is a smaller unit, therefore a larger number than MB or GB counting by units of 1024. This is NOT cheating, it's the same amount of bytes either way, just different units. It is just units, and that is how many there are, and is just how humans count (in powers of 10) - and million IS THE DEFINITION of Mega.

However, after formatting the disk, the computer operating system has notions to count it in binary GB. The device manufacturer did advertise it correctly, and formatting did NOT make the disk smaller, the units just changed (in computer lingo, 1K became counted as 1024 bytes instead of 1000 bytes). This is why we buy a 500 GB disk drive (sold as 1000's, the actual real count, the decimal way humans count), and it does mean 500,000,000,000 bytes, and we do get them all. But then we format it, and then we see about 465 gigabytes of binary file space (using 1024). All precisely correct, 500 GB / (1.024 x 1.024 x 1.024) = 465.661 GB. But users who don't understand this switch assume the disk manufacturer cheated them somehow. Instead, no, the disk just counted in decimal, same way as we humans do. No crime in that, mega does mean million, and we do count in decimal (powers of 10 instead of 2). It is the operating system that confuses us, calling it something different.

However, Memory chips (also including SSD and Compact Flash and SD cards and USB flash sticks, which are all memory chips) are different, and construction requires the use binary kilobytes (1024) or megabytes (1024x1024) or gigabytes (1024x1024x1024). This is because each added address line exactly doubles size, 8 bits is 256, 9 bits is 512, 10 bits is 1024. But also (for no good reason) file sizes are usually said in binary 1024K units. Doing this for file sizes is debatable, but there are good necessary technical reasons for memory chips to use binary numbers, because each address bit is a power of two - the sequence 1,2,4,8,16,32,64,128,256,512,1024... makes it be extremely impractical (unthinkable) to build a 1000 byte memory chip. It simply would not come out even. The binary address lines count to 1024, so it is necessary to add the other 24 bytes to fill it up. But there is no good reason today for file sizes in binary today, it is just an unnecessary complication, however counting in binary 1024 units is still done on them. If we have a file of actual size 20,001 bytes, the operating system will call it 19.532 KB.

Units of 1000 are extremely handy for humans, we can convert KB and MB and GB in our head, by simply moving the decimal point. Units of 1024 are not so easy, but it came about in the computer early days back when 1024 bytes was a pretty large chip. We had to count bytes precisely to ensure the data would fit into the chip, and the 1024 number was quite important for programmers. Not still true today, chips are huge, and exact counts are unimportant now. Hard drives dimension size in units of 1000, but our operating systems still like to convert file sizes to 1024 units. There is no valid reason why today...

As a computer programmer, back in the day decades ago, I had the job of modifying a boot loader in a 256 byte PROM. It was used in 8080 test stations that booted from a console cassette tape, and I had to add booting from a central computer disk if it was present. I added the code, but it was too large. After my best tries, it was still 257 bytes, simply one byte too large to fit. It took some time and dirty tricks to make it fit. Memory size was very important then, but today, our computers have several GB of memory and disk, and the precise data sizes really matter little. Interesting color, at least for me. :)

The definition of the unit prefix "Mega" absolutely has always meant millions (decimal factors of 1000x1000) - and of course it still does mean 1000, it does NOT mean 1024. However, memory chips are necessarily dimensioned in binary units (factors of 1024), and they simply incorrectly appropriated the terms kilo and mega, years ago... so that's special, but we do use it that way. In the early days, when memory chips were tiny, it was useful to think of file sizes in binary, when they had to fit. Since then though, chips have become huge, and we don't sweat a few bytes now.

And with the goal to preserve the actual decimal meanings of Mega and Kilo, new SI units Ki and Mi and Gi were defined for the binary powers of 1024, but they seem ignored, they have not caught on. So, this still complicates things today. Memory chips are binary of course, but there is absolutely no reason why our computer operating system still does this, regarding file sizes. Humans count in decimal powers of 10.

Note that you will see different numbers in different units for the same file size dimension:

Scanning any 6x4 inch photo will occupy the amounts of memory shown in the table below. I hope you realize that extreme resolution rapidly becomes impossible.

You may enter another resolution and scan size here, and it will also be calculated on the last line of the chart below. Seeing a result of NaN means that some input was Not a Number.

Scan size: by inches cm

At scan resolution: dpi    

(Result is shown on last row of the table below)

When people ask how to fix memory errors when scanning photos or documents at 9600 dpi, the answer is "don't do that" if you don't have 8 gigabytes of memory, and a 9600 dpi scanner, and have a special reason. It is normally correct to scan at 300 dpi to reprint at original size (600 dpi can help line art scans, but normally not if color or grayscale photos).

Notice that when you increase resolution, the size formula above multiplies the memory cost by that resolution number twice, in both width and height. The memory cost for an image increases as the square of the resolution. The square of say 300 dpi is a pretty large number (more than double the square of 200).

Scan resolution and print resolution are two very different things. The idea is that we might scan about 1x1 inch of film at say 2400 dpi, and then print it 8x size at 300 dpi at 8x8 inches. We always want to print photos at about 300 dpi, greater scan resolution is only for enlargement purposes.
The enlargement factor is Scanning resolution / printing resolution. A scan at 600 dpi will print 2x size at 300 dpi.
Emphasizing, unless it is small film to be enlarged, you do not want a high resolution scan of letter size paper. You may want a 300 dpi scan to reprint it at original size.

When we double the scan resolution, memory cost goes up 4 times. Multiply resolution by 3 and the memory cost increases 9 times, etc. So this seems a very clear argument to use only the amount of resolution we actually need to improve the image results for the job purpose. More than that is waste. It's often even painful. Well, virtual pain.  <grin>


Copyright © 1997-2017 by Wayne Fulton - All rights are reserved.

Previous Main Next