While it has long been appreciated that imaging biological samples by cryo-electron microscopy is severely limited by electron beam damage, only recently has it become clear that in vivo imaging by fluorescence microscopy is equally sensitive to photon dose, seriously limiting either the density of time sampling either duration of data acquisition, or the illumination intensity, and leading to weak fluorescence signal. It turns out that biological data are completely unrecognizable in the face of the noise that dominates the images (e.g. Poisson noise from limited photon counts plus Gaussian noise from the camera). While the maximum tolerable dose may represent a fundamental limit for the biology, the question is how much information can be extracted from images obtained at the limited dose. Fortunately, new computational approaches and algorithms are beginning to show promises. In this talk, we will focus on patch-based methods already investigated in [1-5] that utilize local intensity statistics in 2D or 3D to find patches of similar behavior, replacing each pixel/voxel with an appropriate weighted sum. We propose several extensions of the existing algorithms and present new investigations in cryo-electron tomography. The performances of the algorithms are evaluated on both synthetic and real image sequences using qualitative and quantitative criteria.