Per-pixel background estimation in video microscopy using frame grouping and wavelet based image fusion

Per-pixel background estimation is very important for accurate tracking of biological structures in video microscopy. Mixture modeling of the background is hard due to less priors and heavy halo effects around the target biological structures in each frame. Since halos could have similar brightness and motion pattern to the foreground sometimes they are not identified as in the background by these methods. In this paper we proposed a new method based on incremental frame grouping and iterate image fusion to compute the per-pixel background in video microscopy. Experiments on simulated microscopy videos and real sequence data set show that our method could calculate the background with a quite high precision. At the same time it also outperforms previous methods in removing strong noise and heavy image blurring, the “halo”, which are especially useful attributes for video microscopies.