How I Did My Image Processing
Simple steps that will allow you to do it yourself - part 4
Suppose that I have a fairly clear image, but I am not happy with the poor level of detail in it. There are some things that can be done to enhance details, but care must be taken- every gain comes at a cost.
|
Let's start with a picture that has some subtle forms but the details are lacking. We can barely make out something, or maybe we can see it fairly well, but somebody else just doesn't see it at all. What can be done? We can start by making a copy of this image. It will be our work image. Think about the picture and the data that is in it. We know that there is image data that we can see, and lots more that we cannot. That is because our eyes work best in a narrow range of contrasts and brightnesses. So, let's imagine an image that has only the changes from light to dark that represent these details, minus the bright spots and shadows that are present in any real-world image. Can we get that data out so we can see it directly? |
 |
|
Actually, that is a very simple process. Picture the creation of a perfect "anti-shadow" that would erase the light and dark information and leave only the change data- whether something was getting lighter or darker, not whether it is light or dark in this picture. Since this hypothetical perfect shadow would have no fine details in it, let's see if we can create it from the original image. We take our copied work image and we blur it a bit like this. You will notice that, oddly, blurring the picture actually makes some features seem to blend together and become a bit easier to see. Technically, we are dealing with something called "spatial frequency". If we blur an image we are destroying the high spatial frequency data, and leaving the "blurry" low frequency spatial data. This is exactly what we want. |
 |
|
| Now that we have
this blurry picture, we want to subtract it from the good image. The
simplest method is to invert the image, like a photographic negative.
This is so far not very different from the previous "edge enhancement" method I took you through. But there are major differences- this method will not cause an effect called "ringing", where small details could appear to duplicate or echo through the image. In fact, this method is simpler and cleaner and can usually be used anywhere the first method can. It can also be done many times, with various levels of blurring, to extract the features in various spatial frequency bands. If this is a bit technical sounding, don't worry. The method works whether you understand it or not. |
 |
|
| The next step is
to add this negative shadow (that anti-shadow I mentioned) to the original
picture. We will arrive at a strange looking result that has only
the details of the image, not the actual illumination data.
This image can then be contrast-enhanced to show fine features that we could not see originally. Our eyes just are not adapted to find such features across the range of light and dark that we can perceive all at once. Here is the high-frequency difference map that will come out of this process. One thing you will notice is that there appears to be a lot of JPEG noise in this image. Well, that is the consequence of the image compression method that NASA is using. We will have to sacrifice a little detail to get rid of it, by doing a deliberate low-level blur to the image. Don't worry, the end results will still be very good. |
 |