User talk:Dawoodmajoka
From Wikipedia, the free encyclopedia
Hi, I noticed that you have written some stuff on motion estimation: Horn Schunck method and Lucas Kanade method. Can you please motivate why they both are described in terms of 3 "spatial coordinates" (x,y,z) and time t? Both methods can be generalized in this fashion and I guess that there may be some application on 3D+time data where this generalization is useful. But neither of the original publications of these methods nor any of the standard presentations on these matters which I consulted describe these methods in this general way. As an introduction to these methods, the current presentations appear a bit complicated since the reader cannot related to optical flow in a 2D image as a function of time. --KYN 19:33, 17 March 2006 (UTC)
Sorry for the very late reply, I did not check in for long time.
The 3D+time applications are exactly the area where I have used the algorithms. Its simple to convert any of them to the 2-D case by omitting one dimension. The specific area where I have used them is medical imaging. We had 3D images from patients taken over a long period of time and they were to be transformed to a single point in time. You can check my article in IEEE Transactions on medical imaging, 2006, special issue on pulmonary imaging.
Another article is currently under review. User:dawoodmajoka
OK, I see. It's perfectly clear to me that the type of extension which you describe here is possible (and straightforward) and also relevant for certain applications. However, as I said previously, neither the original publications nor standard presentations on this matter deal with 3D data which varies with time. I suggest that it is better to modify the articles such that they are based on 2D+time data and insert a section which describes the generalizations which you are interested in. Given the context of the standard case, this can be done fairly straightforward. --KYN 18:34, 8 August 2007 (UTC)
