Six Million Pixels

You are about to be assaulted by a new stream of lies from marketing departments. When you buy a digital camera you will see 6 Million Pixel File plastered all over cameras that cannot capture 6 Million pixels of image. This article explains the new fraud headed your way. Unlike the fraud at Enron, this form of fraud appears to be legal.

I am looking at an advertisement for Fijifilm digital cameras. The description for the CAM-S602 says 6 Million Pixel File and 3.1 Million Effective Pixels. When you read marketing speak about capacity and there are two numbers, the larger number is usually the lie and the smaller number an indication of truth. If there is one number in big print at the top of the advertisement and a different number in fine print at the bottom of the page, that is a good indication the number in big print is a lie.

The file size fraud is used with a number of models in the Fujifilm advertisement. The example I mention here has a 3 Mp camera misrepresented as a 6 Mp camera. The next step will be to misrepresent a cheaper 2 Mp camera as a 6 Mp camera because digitally smearing 2 Mp of pixels over 6 Mp is as easy as smearing 3 Mp over 6 Mp. In fact any toy manufacturer could get one of those sub Mp toy cameras and smear the already blurry image over 12 Mp then claim to have the cheapest camera in existence that can produce a 12 Mp image.

Spreading 3 Mp, or 0.3 Mp, over 6 Mp is done through interpolation and there are many formulas for interpolation. All the formulas are designed to fake a certain type of image better than other types. One formula might do a good job of representing geometric images at large sizes while another image might perform a better job with continuos tones. No matter which formula the camera maker uses to fake a larger file size, no formula can produce an accurate file because no formula knows the missing information.

The following exaggerated example uses images 100 pixels by 100 pixels. One image has a vertical line and the other a diagonal line. An interpolation formula that fakes a larger file size by multiplying pixels will perform a good job on the vertical line but make a horrible mess on the diagonal line. A formula than averages pixels will produce an equally unsatisfactory result with both.

Sample Files

The following image is a sample with the top left half coloured white and the bottom right half coloured black. The colours form a line from the top right corner to the bottom left corner. You will see the effect of resizing and interpolation along the edge of the black section.

The following image has the left side coloured white and the right side coloured black. The colours form a line down the middle from top to bottom. You will see the effect of resizing and interpolation along the edge of the black section.

Files from a Cheap Camera

The following sample images are the previous images reduced to one-tenth their original size. The small size represents part of a small image from a cheap digital camera. The same image also represents part of the small image produced by the image sensor in fake cameras before the cameras apply their fakery.

We will expand these images using the same techniques some marketing people and some camera makers use to inflate the resolution of their products.

Expansion by Pixel Multiplication

Here is the small vertical line image expanded back up to 100 pixels by 100 pixels using a simple pixel multiplication formula. Notice the line has a clean edge. Pixel multiplication formulas work well with clean vertical and horizontal lines.

Here is the small diagonal line image expanded back up to 100 pixels by 100 pixels using a simple pixel multiplication formula. Notice the line is a ragged, jagged edge looking completely different to the original clean line.

Expansion by Pixel Averaging

Here is the small vertical line image expanded back up to 100 pixels by 100 pixels using a pixel averaging formula. Notice the line is a nasty blurred edge that looks completely different to the original clean line.

Here is the small diagonal line image expanded back up to 100 pixels by 100 pixels using a pixel averaging formula. Notice the line is a nasty blurred edge that looks completely different to the original clean line. The ends of the line are further damaged because averaging fails when there are few points to average.

The following image is the small diagonal line image expanded back up to 100 pixels by 100 pixels using a pixel averaging formula then multiplied to 500 pixels by 500 pixels to highlight the damage created by the previous inflation. The line is a nasty blurred edge that looks completely different to the original clean line. The ends of the line are further damaged and there are blotches of grey in the white area. The image inflation routine attempted to make the white and black areas visually balanced by slapping extra distortion in the white area.

Mathematics is Not Knowledge

Image expansion programs make all sorts of guesses about what you want to see. The software developers assume you use exactly the same monitor as them. Some assume you are stupid enough to blame the poor quality on something other than the camera.

To correctly work out what should be in your 6 Mp image, the only thing the software can do is read a 6 Mp image from the image sensor in the camera. If the image sensor produces anything less than 6 Mp then the software cannot calculate the missing information. No amount of fancy mathematics will replace the missing image information.

When You Really Want a Larger File

There may be times when a larger image is useful for presentation. The thing to remember is that every change of image size introduces damage so keep the number of changes to the minimum. Get a camera that outputs exactly what the image sensor senses. Store the exact output on disk and never change the original. When the image is needed for presentation, make a copy of the image and resize the copy. Change the image size in one step to the exact size needed for presentation. If the size is wrong, make a new copy from the original and resize the new copy.

If you have to experiment with several size changes, you can make all the size changes on one copy, add up the changes then reapply the changes as one to a new copy. The PaintShop Pro image editing software lets you apply size changes as a percentage of size or by specifying the final pixel size. Using PaintShop Pro, you could experiment with percentage size changes until the size is perfect, record the final pixel size, copy the original file again, and specify the recorded pixel size as the final size.

Damage Is Damage Even If It Is Not Immediately Visible

I sometimes have to work with images that have suffered several small size changes. The result is incredibly blurry. Even though each step was a minor change to the person editing the image, the cumulative damage looks like a jellyfish died on top of the image. Imagine how difficult life becomes if the camera damages the image before the first human gets to make a decision about quality.

Skin cancer is caused by cumulative invisible to your skin. First you get freckles then moles then you skin drops off then you die. Each step takes years and on a day to day basis, you never see the change. The same happens when you let software modify an image without your permission. The end result is damage no matter how hard it is to see each form of damage.

Conclusion

One 3 Mp camera is not worth more than another 3 Mp camera just because the camera can produce a fake resolution. The fake 6 Mp image is worth less than a real 3 Mp image because you cannot remove the damage done by the interpolation. That makes the Fijifilm camera, and any other camera that produces fake resolutions, worth a lot less than truthful competitors.