Denoising of Microscopy Images: A Review of the State-of-the-Art, and a New Sparsity-Based Method

Abstract

This paper reviews the state-of-the-art in denoising methods for biological microscopy images and introduces a new and original sparsity-based algorithm. The proposed method combines total variation (TV) spatial regularization, enhancement of low-frequency information, and aggregation of sparse estimators and is able to handle simple and complex types of noise (Gaussian, Poisson, and mixed), without any a priori model and with a single set of parameter values. An extended comparison is also presented, that evaluates the denoising performance of the thirteen (including ours) state-of-the-art denoising methods specifically designed to handle the different types of noises found in bioimaging. Quantitative and qualitative results on synthetic and real images show that the proposed method outperforms the other ones on the majority of the tested scenarios.

A Poisson-Gaussian Denoising Dataset With Real Fluorescence Microscopy Images

“…Specif-ically, since Poisson-Gaussian noise is a mixture of both Poisson and Gaussian noises, which are parameterized by a and b, respectively, an image with a large estimate value of a but a small b may be considered as a Poisson noise dominated image, while a small a with a large b can indicate that the image is Gaussian noise dominated. In fluorescence microscopy, however, it is unlikely to have a Gaussian noise dominated image due to the low signal levels; most fluorescence microscopy images are Poisson noise, or shot noise, dominated, with certain types of microscopes, such as widefield ones, have a considerable amount of Gaussian noise involved [5,20]. Note that the noise estimation program from [10] could generate a negative b value when the Gaussian noise component is small relative to the pedestal level (offset-from-zero of output).…”

HISTOBREAST, a collection of brightfield microscopy images of Haematoxylin and Eosin stained breast tissue

“…Image enhancement techniques aim at emphasizing particular features of interest or at making images more visually appealing and, more importantly, easier to interpret, without utilizing any image formation model. They typically include image denoising 44 , contrast enhancement 45,46 and flat-field correction methods 47,48 . The HSIMs available in HISTOBREAST, assembled from NITs systematically acquired under various acquisition settings, can be used for the development and benchmarking of image enhancement algorithms suitable for brightfield microscopy and histopathology.…”

A bird’s-eye view of deep learning in bioimage analysis

“…Automated bioimage analysis typically requires executing an intricate series of operations, which may involve image restoration [10] , [11] , [12] and registration [13] , [14] , [15] , object detection [16] , [17] , [18] , segmentation [17] , [19] , [20] , and tracking [21] , [22] , [23] , as well as downstream image or object classification [24] , [25] , [26] , quantification [27] , [28] , [29] , and visualization [30] , [31] , [32] . As attested by the just cited reviews and evaluations, a plethora of methods and tools have been developed for this purpose in the first half a century of computational bioimage analysis, based on what may now be considered traditional image processing and computer vision paradigms.…”