An FPGA-based design for a real-time image denoising using approximated fractional integrator (Q2211681)

A new algorithm proposed to remove noise, especially impulsive noise, from digital images while preserving the essential information about the scene captured in the image is discussed in this paper. This method is based through the use of differential operators of fractional order applied to the image function. Let's see a brief introduction to the mathematics of this paper. There are several definitions of fractional order derivatives. In this paper, the authors adopt the following one. The image function is $f:\mathbb{R}\to\mathbb{R}$ such that $f(x)=0$ if $x\notin[a,b]$ where $-\infty<a<b<+\infty$. The $q$-order fractional operator of \(f\) in $x\in[a,b]$ is (if $q>0$) \[ _{a}D_{b}^{q}f(x)=\lim_{h\to 0}(1/(h^{q}))\sum_{i=0}^{q}(-1)^{i}(q/i)f(x-ih), \] (if \(-1<q<0\)) \[ \Delta^{q}f(x)=(1/(\Gamma(-q)))\lim_{h\to 0}\sum_{i=0}^{\infty}((\Gamma(i-q))/(\Gamma(i+1)))f(x-ih). \] The image with noise removed is $_{a}D_{b}^{q}f$ or $\Delta^{q}f$. Simulation results of the application of the proposed method on black \& white and grayscale images are shown in Section 3. Also by simulation, the authors compare the proposed method with several other ``classical'' and recent methods. To do so, different parameters are applied, some of them well known such as the correlation coefficient between the pure image and the images resulting from the application of the respective method, some others proposed in previous papers by other authors, and others proposed in the paper. Regarding new noise removal techniques, proposed from 2015 onwards, the authors acknowledge that in a black \& white image, a few of them were superior to the proposed one, while others gave comparable or inferior results. The proposed method is generally said in the paper to be satisfactory. One section of the paper is reserved for describing hardware implementation for the proposed algorithm on a particular device. The list of references is extensive and up to date.