I am working on some format conversion routines for video processing and am wondering if anyone has experience with multiple conversions and subdueing some of the conversion errors with this. The purpose is for a library of video processing functions and convenience functions that handle capture, X display and so forth. Some of the later algorithms may use any of RGB, YIQ, XYZ or HSV formats, so in checking them, I have converted to one and then back, converted to one, then to another, then back and so forth. All of the basic conversions are quite invertible and look nice without problem pixels from the transformations, but many of them have problems with getting back to the original image when converted between several formats in succesion. Normalizing the images to their bounds does not help, nor does clipping them to their bounds. It is somewhat moot that this work flawlessly since there are few applications where one would be doing so much converting, but I still wonder if anyone has any ideas. My thoughts are that in some of the conversions that are posed as linear algebra problems, they may be somewhat poorly conditioned and in using several one gets enough error from the original that the image develops problem areas for some values. As soon as I get some more done on the library, I will stick it up somewhere for general availability so interested parties might fool with it. Just checking if anyone has had this problem and came to some viable solution. I haven't yet really worked on this issue too much, so perhaps there is an obvious correction or compensation that I haven't thought of yet, but I still thought I would ask. Best Regards, CS
