Image Denoising Using Backward Stochastic Differential Equations

Reduction of JPEG Artifacts using BSDEs

“…In this section we present experimental results illustrating the difference between our algorithm and other methods of removing artifacts: SADCT The analysis of the measures of image quality from the Table 2 shows that the new method performs better, especially in the case of low value of JPEG quality. In particular, our modification of the algorithm by adding weights improves results of the parent method [Bor17].…”

“…Now, we can define the process Y . From a definition of BSDE [Bor17] it starts from Y t m = u 0 (X t j ) and then we backwardly count values…”

“…In this paper we present a new method of artifacts reduction in the case of RGB images. We combine the idea of image denoising based on backward stochastic differential equations from [Bor17] and the concept of applying the non local means method to Feynman-Kac formula from the paper [Bor14].…”

“…The rest of the paper is constructed as follows. In Section 2 we recall the method of denoising in terms of backward stochastic differential equations from [Bor17]. Section 3 provides a new method of restoration of the noisy image.…”

“…[Foi07], CBM3D[Dab07], ARCNN[Don15], DNCNN[Zha17] and BSDE[Bor17] as the parent method for our approach. We use the MATLAB implementation of compared methods: SADCT[Foi20b], CBM3D[Foi20a], ARCNN[Yuk20], DNCNN[Jpe20].…”