Adaptive Cloning (aka Seamless cloning)
Direct cloning of parts from one image to another, usually ends in bad results because of different lighting conditions and other settings (such as white-balance) which causes the color of the cloned part not to match the source image and look out of place. There are some techniques to solve this, by using poisson equations and some other methods. This suggestion allows a fast (realtime) solution for allowing to paste seamlessly parts from one image to another.
The project I intend to implement is something called Adaptive Cloning (aka Seamless cloning). Basically, the idea is that cloning parts from one image to another usually does not end well. This is because of different exposure/white-balance settings, or simply because of different lighting conditions.
The idea is to implement an algorithm which would allow to copy a part from one image, and paste it seamlessly into the other; the pasted area will be blended to fit the destination image locally. There are several techniques to do that, and the most known one is called poisson image cloning. The problem with most of these algorithms is that they are expensive (computation-wise) and are not suitable for interactive editing and previewing, unless you use hardware acceleration.
The suggested technique for my project is based on an article from Siggraph 2009 (includes demos and videos), which shows realtime performance (both on the GPU and the CPU*). The user would paste a some area from the clipboard and will then be able to "blend it in" according to it's environment.
I'm open for any suggestions regarding usability, ui integration, or anything which relates to the project. Feedback will be more than welcome!
* Note that I'm not sure for which sizes on the CPU it preforms on realtime, but it's still much more cheap computation-wise than most other techniques and a small lag of few seconds is something which should be acceptable for this sort of workflow.
|File name||Size||Date submitted|