A Space-Variant Nonlinear Algorithm for Denoising Omnidirectional Images Corrupted by Poisson Noise

An omnidirectional camera uses a conventional camera in conjunction with a quadratic mirror for capturing a 360° field of view in real-time. The resolution of an omnidirectional image, however, is non-uniform due to the mirror curvature. It can lead to the degradation of performance of classical methods when applied directly to omnidirectional images. In this paper, we investigate an adaptive algorithm for denoising omnidirectional images corrupted by Poisson noise. We propose a space-variant regularization based on total variation, which is adaptive with the non-uniform resolution of omnidirectional images. The weighted functions based on the mean curvature of an image surface are formulated to adapt the fidelity and the regularization terms to fit the data and preserve fine details. Experimental results demonstrate that the proposed method can better remove noise while maintaining features in comparison with the state-of-the-art methods.