Normalization (as opposed to calibration)

I am not sure I am using the right lexicon but here is my issue. End user has a series of images taken with the same sensor but at different days and times. They have asked me to write a piece of software to "calibrate" the images such that they can compare them and detect changes over time. I think from listening to their requirements that while the sensor images may need calibration, what they are really after is a way to normalize all images to some constant. They believe that if they can give me a 'feature' that exists in all images I should be able to determine the pixel values of that area, compare it to each subsequent image (in the area of that same feature) and then normalize all images to some median value.

My question is whether this is a capability of the LT toolsuite and if so, which one? It is a little out of scope of what I normally do...

Thanks,
-DR

Depending on the type of difference between images, our toolkit might have useful functions for your requirements. Please answer the following questions:
1. If I understand correctly, all images will contain the same background (same scene) taken from the same angle and at the same zoom level, but using different lighting conditions. Is this correct?
2. What other differences might there be? In other words, what is the purpose of the comparison? Is it to detect objects or movement?
3. Are there sample images that show the input and output of the required process? If yes, please send me these samples in a ZIP or RAR file to ensure they are delivered correctly.
You can either post it here or send it to support@leadtools.com and mention this post.
When attaching to the forums, do not use the "Preview" feature.

I will try. Thanks for your help.

1) The lighting is the major change although I can't say that the angle will be EXACTLY the same. We can crop and zoom to take those attributes out of the equation.

2) The object of the comparison is both to detect new features, the absence of features or displaced (moved) features. Probably normally it would be the introduction of a new feature.

3) I really wish I had good samples to send but I don't have release capability on the stuff I have at this time. I am trying to get releasable images.

Thanks again.

The implementation will depend mainly on what types of differences exist between the images. That's why it's important to have sample images from the actual application environment.
Here are the main ideas:
1. For the calibration step, a common area should be defined accurately and the average of all pixel intensities is calculated. Then the whole image intensity is increased or decreased accordingly.
2. The locations of image parts should the same to accomplish accurate comparison. If small shifts can occur from one shot to the other, corrections should be made. One way to do that is to look for known objects (using the correlation function) then triangulate the amount of shift and/or rotation from a standard template image. This technique is called Registration Marks in LEADTOOLS.
3. Once the 2 images to be compared have the same location and approximate intensity values, they can be subtracted from each other using the Combine function. Since there would be small differences, some noise reduction features could be used to keep only actual object differences.

If you want to discuss the details further and send sample images, please specify which programming interface (.NET classes, DLL API, C++ Classes, etc.) you want to use.