Images and Pixels

Images and Pixels:

Digital imaging devices such as those found in film scanners, print scanners, and digital cameras record a scene by dividing it up into an array of cells called pixels. In a given digital scanner or camera the pixels are all of uniform size and aspect ratio. The dimensions of the pixel array (numbers of pixels high by number of pixels wide) along with the size and aspect ratio of the pixels are generally determined by physical characteristics of the scanner or camera and are not changeable by the user. The image below shows, in a very course, exaggerated form, how a scene is divided into individual pixels, in an array, by a digital imaging device.

How is it then that cameras and scanners offer users the capability to choose the pixel array dimensions for the image they obtain? To start with, as said in other words earlier, the imaging sub-system of the scanner or camera has certain physical attributes that fix a certain pixel array. When the number of pixels desired is different than the basic physical capability of the scanner or camera there are several methods that can be used to get the desired result. Two common methods are as follows: If the image desired is to have less pixels than the basic capability of the scanner or camera some of the data from the imaging device in the camera or scanner is ignored and not saved. If the desired image is to have more pixels than the basic capability of the camera or scanner a math algorithm is applied to the image data to add the extra pixels. In this case the algorithm looks at pixels near where it wants to put a new pixel and makes a "guess" as to the correct red, green, and blue values for the pixel being added.

Sorry, but there isn’t any magic here and the image data that’s being added didn’t exist in your original scene. The more pixels that are added the more obvious it becomes that the new pixels weren’t part of your original scene. So what can be learned from this? Use your digital scanner or camera at it’s basic capability. That way you get all the scene data it can give you. Save the images in that form and if your final desired image needs to be larger or smaller, make the changes in an image editor. You’ll have more control there over what is done and how it’s done and you’ll still have your original to go back to if something goes wrong.

Now there are always exceptions, but in general with a digital camera the basic capability of the camera is most often the maximum pixel array size the camera can produce. Camera’s usually offer "down sizes" so you can save space on your storage media such as flash cards. Scanners are usually the reverse. A scanner’s basic capability is usually the smallest pixel array size offered. This is often put in terms of the scanner’s "optical resolution".

Each pixel in an image can only have one color value; that is one set of red, green, and blue values to represent it. So what happens if in a scene there is more that one color or tone in the area covered by a single pixel? The values for a given pixel are the average color and tone for the light coming from that point in the scene. Now I said point; that could lead you to believe that the pixels are always small so any errors in them are insignificant. Well, that depends on what you want to see in the scene and how much you may intend to enlarge a scene in hopes of seeing some part of it better.

The two images below are a gross demonstration of this point the image on the left shows how the scene will be viewed by the pixels in our, course, example digital imaging device. Each cell in the scene as bordered by the black lines is one pixel. The image on the right shows how the pixels would be recorded as the average value of everything in a given pixel’s area. As you can see there is potential for serious image data loss.

Now this seems a little silly since this ratio of pixel size to image size is not commonly encountered, or is it? Let’s say you just took a picture of a car with your digital camera from several feet away. You need to read the license number on the car’s license plate so you put the image into your image editor and start enlarging it. What you find may indeed look very much like the image above.

Have you noticed in some digital images that areas of the same or similar tone and color, like the sky or a wall, look like they were "painted on sand"? Have you noticed that some digital images seem to have a strong, underlying, texture or pattern? There are several causes for such anomalies. I'll cover a couple of the more common causes.

To begin with, if the image came from scanning a print, look at your print. Problems with a pattern in your digital image can be caused by the actual structure of the paper that the image was originally printed on. Some papers are prepared with a texture that looks nice with the original print, but present problems with digital scanning. Other papers such as those commonly used in sales brochures have a course grain. Again this grain structure can and often does show up in digital scans of prints. If the print is an original photographic print where there was a problem with exposure of either the original scene or of the paper during printing similar effects can be seen. In these cases, sometimes, with a lot of patient work and a good image editor a measure of the undesired effect can be removed. Some such editors offer filters such as "Despeckle" that help remove this sort of problem, but since the filters that are published with a piece of software have to be written for some sort of general case problem they may not be effective with your particular problem. If you are good with algorithm math and can program you could try writing a filter of your own to address your specific problem. Many image editors allow things like filters to be added by their "Plug-in" interface.

The finer texturing of an image such that they look like they were "painted on sand" commonly comes from noise. What's noise? Undesired, random variation in the image data. This is largely caused by the electronics that capture the image and algorithms that are used to process the image data either in the image capture device (camera or scanner) or your computer.

There are usually several places in the circuitry of a scanner or digital camera where electrical noise in the circuit can vary the data values that are recorded for pixels. In the commonly available systems one data byte is used to record each of the three colors (red, green, and blue). As I've mentioned before this results in three bytes of data for each pixel. Each byte is, in turn, made up of 8 data bits. Since you have three bytes of data there are a total of 24 bits (8bits X 3bytes) that represent the data for each pixel. This is where the term "24 bit color" comes from. In these system, because of the variation caused by noise, the state of at least one and sometimes two bits of each data byte are random. So in effect the 24 bit color system has between 18 (6bits of good data X 3bytes) and 21 (7bits of good data X 3bytes) useful bits of data about the image. Enter the 32 bit color system: the goal with this new system is to get 24 bits of good data allowing for the fact there will never be an affordable noise free scanner or digital camera. Yes, 32 bit devices are more expensive than 24 bit devices, but not as expensive as making a noise free device. An easy way to see this noise is to photograph or scan a neutral gray tile. Such things are available for this purpose. These tiles are very flat, very uniform in their gray color, and expensive. You can get an idea though just by scanning or photographing a gray sheet of paper. Put the resulting digital image into your image editor and I don't think you'll have to enlarge the image to much to see that the image is not exactly uniform gray. My print scanner is a very inexpensive 24 bit model and I don't own one of the fancy gray tiles, but I have a gray image I use for such purposes. I've scanned the image and the picture below is a section from a 10X magnification of the resulting scanned image. Sure, no one, well almost no one, looks at there pictures under a microscope, but lots of people like to make enlargements of their pictures. To be fair some of the noise in this sample comes from the fact that it is stored as a .JPG file.

The links at the bottom of your screen will take you to the main topic areas