I have implemented Background remove functionality(aka : Green Screen Implemetation) using kinect in my Windows-RT application over there the noise of pixels (Jitter) is very high at foot area as well on hair of the acquired user so how to reduce this noise of pixels ?
There are a few techniques you could apply to reduce noise:
cv::bilateralFilter, most intensive, but with the right number of iterations will smooth out the image.
cv::morphologyEx, morphological closing will remove small gaps (of a few pixels) in the image, if the structuing element (cross, circle etc.) is the right kind and size.
cv::inpaint, will close bigger gaps and fill out the image where data is unavaliable. I suggest trying bilateral filtering (smoothing) after this step.
cv::findContours, filtering contours with an area smaller than a threshold could be used to remove big gaps in the image.
1 & 3 are mostly for salt and pepper noise and 2 & 3 are most appropriate in removing missing data.
Scaling down the depth data and scaling it back up to size (with good interpolation) also has the effect of smoothing out the image whilst preserving edges.
Using the K2, you might also find that mapping from color to depth coordinate space or vice vera gives you better results than the former.
Lastly, I would suggest you to look at some techniques used by traditional green screening and VR/AR, such as colouring the outermost edges of the foreground with a light or dark outline to get a 'clean' look.