Anisotropically foveated nonlocal image denoising

Abstract

When our gaze fixates a point, the visual acuity is maximal at the fixation point (imaged by the fovea, i.e. the central part of the retina) and decreases rapidly towards the periphery of the visual field. This phenomenon is known as foveated vision or foveated imaging. We recently investigated the role o ff o v a t i o ni ni m a g efi ltering and we have shown that the foveated patch distance, i.e. the Euclidean distance between foveated patches, is a valuable feature for the assessment of nonlocal self-similarity. Foveation operators apply spatially variant blur, providing a compact multiscale representation of each image patch. Here, we introduce anisotropic foveation operators that embed directional point-spread functions, and we show that the operators providing the highest denoising quality are characterized by radial orientations. This result is coherent with the orientation preference in the human visual system.

Improved Microaneurysm Detection using Deep Neural Networks

“…Visual acuity is maximum in the middle of the retina termed as fovea and decreases towards the periphery of the retina. Foveation proved to be very effectively in nonlocal means denoising algorithms [21]. In foveation central section of the window is focused, while the peripheral pixels are defocused using linear space invariant gaussian blur.…”

Foveated Nonlocal Self-Similarity

“…Thus, foveated self-similarity reveals a connection between features of the HVS and modern imaging algorithms. This work builds upon our preliminary conference publications: Foi and Boracchi (2012), where we first introduced isotropic foveation operators and their constrained designed for Foveated NL-means;and Foi and Boracchi (2013b), Foi and Boracchi (2013a) where we described their anisotropic extensions. Here, we develop these works in the following directions: we provide a more precise and detailed characterization of the design constraints leading to foveation operators (both isotropic and anisotropic); introduce selfmap foveation operators that share the same compact support of windowing operators; illustrate the foveation operators through their singular value decomposition; present a more thorough and extensive experimental validation which includes a detailed analysis and optimization of the filter parameters; detail the computational overhead of replacing the windowed distance with the foveated distance; discuss several connections between the HVS and the effectiveness of foveated self-similarity, including a direct analogy between the radial bias in the HVS and the optimal orientation in anisotropic foveation, which we expose by replicating the outcomes of a recent fMRI study on human subjects within our image-denoising framework.…”

FPGA Based Optimized Reconfigurable Base-2 Constant Coefficient Multiplier Architecture for Image Filtering

“…[10]- [11]. The digital convolution [2]- [3] process widely implemented in FPGA [4] for several decades. However these algorithms are computationally expensive.…”

“…But the multiplier design is generic and not suitable for filtering [3]. A.G.Dimster suggested the Fixed binary common sub expression algorithm (FBCSE) is eliminate the computational complexity moderately by considering BCSE algorithm across the adjacent coefficients is designed [4], [9],[10].…”