EN FR
EN FR




Bibliography




Bibliography


Section: New Results

Noise modeling and denoising for intensified camera in fluorescence imaging

Participants : Philippe Roudot, Charles Kervrann.

Two papers under review.

Figure 9. Variance prediction after noise model calibration using a fluorescein FD-FLIM reference stack acquired with a wide-field (WF) microscope (parametric model in red and measurements in blue).
IMG/FLIM-global_2_crop.png
Figure 10. Denoising performance on a live cell image acquired in FD-FLIM (fluorescent tagged caveolin protein) using a confocal setup. From left to right: original image, results with bm3d [27] nD-Safir [4] and our method.
IMG/LiveR1Cav1_control_120005.pngIMG/bm3d_00006.pngIMG/denoised_LiveR1Cav1_control_12_f6_2d_bis.pngIMG/FLIM_WF12_0005.png

Image intensifiers are commonly used in low light level biological imaging, especially for fluorescence imaging. In this study, we have proposed a statistical framework for noise variance estimation dedicated to image sequences acquired with ICCD (Intensifier CCD). The model has been exploited for fluorescence lifetime estimation (Fluorescence lifetime imaging microscopy, FLIM) [13] , [12] and image denoising. We have investigated an alternative approach to [41] and we have shown that intensifier gain variation cannot be neglected in the variance estimation as opposed to a CCD sensor gain. Additionally, we have suggested to correct the noise model spatially to cope with microscopical aberration which are common in experimental setups (see Fig. 9 ). Finally, we have proposed a novel denoising algorithm based on the NL-means filter [23] which does rely on variance stabilization. The novel patch-based filter is able to adapt to local intensity-based noise statistics (see Fig. 10 ).

Partners: F. Waharte and J. Boulanger (UMR 144 CNRS PICT IBiSA Institut Curie)