Person Re-Identification in Panoramic Views Based On Bayesian Transformers

This paper explores the problem of removing real, non-simulated noise from a large body of large binary images. Three denoising methods are evaluated for their efficacy and speed: the well known DUDE, a novel variant of it which we call the Quorum Denoiser, and an adaptation of the Non-Local Means (NLM) method for binary images, B-NLM which, to our knowledge, is faster than other known variants. The methods are compared and tested both on simulated noise (as a benchmark) and on the real life images. All three methods produce good results on real noise. However, despite being optimized, the B-NLM is significantly slower than the other two, whose speeds are comparable to a plain median filter. Overall, the Quorum denoiser appears to be the best option, both in quality (real and simulated) and speed.