Denoising imaging polarimetry by adapted BM3D method

Abstract

In addition to the visual information contained in intensity and color, imaging polarimetry allows visual information to be extracted from the polarization of light. However, a major challenge of imaging polarimetry is image degradation due to noise. This paper investigates the mitigation of noise through denoising algorithms and compares existing denoising algorithms with a new method, based on BM3D (Block Matching 3D). This algorithm, Polarization-BM3D (PBM3D), gives visual quality superior to the state of the art across all images and noise standard deviations tested. We show that denoising polarization images using PBM3D allows the degree of polarization to be more accurately calculated by comparing it with spectral polarimetry measurements.

Linear Polarization Demosaicking for Monochrome and Colour Polarization Focal Plane Arrays

“…However, existing methods tend to overestimate DoLP, so in dark regions these methods could have DoLP values larger than 1, which is physically impossible. This has been pointed out by Tibbs et al [TDRB18].…”

“…Systematic evaluation of our demosaicking method and competing techniques requires a realistic, high quality test dataset with ground truth results. To the best of our knowledge, existing polarization image datasets are monochromatic, and consist of only a few scenes [LGFB18,TDRB18]. High-resolution polarization images with colour, are lacking for research.…”

“…However, effective polarization reconstruction for such sensors remains a largely uninvestigated problem. As pointed out by Tibbs et al [TDBR17,TDRB18], noise in the measurement may lead to significant artefacts and incorrect conclusions. We show a simple example in Figure 2b, where Gaussian noise with a standard deviation of two out of 255 is added in the measurement, and apparent artefacts arise on the edges in the Degree of Linear Polarization (DoLP) estimation by bilinear interpolation.…”

Signal denoising of viral particle in wide-field photon scattering parametric images using deep learning

“…This issue becomes even worse when imaging nanoscale samples below the diffraction limit, such as nano-particles or nanopores. In some polarization imaging methods, such as division of focal plane (DoFP) polarimeters, specialized denoising algorithms [17][18][19] are designed, which however cannot be directly applied to PIMI because of their notably different measuring and reconstruction procedures. Besides, deep learning methods [20][21][22][23] are also widely used for image denoising while they typically require a huge training data sets, and it is difficult and impractical to make a sufficient amount of ground truth data in PIMI.…”

TomoGAN: low-dose synchrotron x-ray tomography with generative adversarial networks: discussion

“…Thus, previous approaches have found it beneficial to extend the loss function to be minimized by the GAN generator with a more traditional loss, such as the 2 -norm distance (i.e., mean squared error) [37]. 2 -norm distance is one of the most commonly used loss metrics for image restoration problems [23,8] although it is well known to produce blurry results on image generation problems in classical approaches [38,39]. Thus, we also add 2 -norm loss for the generator to enforce correctness of low-frequency structures [26,40,41].…”