Fast Denoising for Fluorescence Image Sequences in a Nonlocal Means Framework

In fluorescence live-cell imaging there is always a trade-off between image quality and cell viability. While avoiding photo bleaching and photo toxicity, light exposure time must be limited which results in low signal-to-noise ratio. We present a fast non-local means technique to denoise 3D image sequences acquired via fluorescence microscopy. The commonly used non-local means filter for image sequences is computationally inefficient. We reduce the computational cost by carrying out the denoising in the lower dimensional subspace determined by principal component analysis (PCA). Image neighbourhoods are projected onto the lower dimensional subspace determined by PCA. We use shot boundary detection as a preprocessing step to identify and form different shots with content-wise similar frames. We show that the proposed method reduces the computations in addition to improving the accuracy. Our results are also compared with other fast NLM based techniques.