A universal de-noise algorithm with trilateral filter and impulse detector

In this paper, we introduce a new edge-preserving nonlinear filter for removing the mix of Gaussian and impulse noise. Built from Prasun Choudhury and Jack Tumblin's trilateral filter, the new algorithm incorporate a local gradient statistic for detecting corrupted pixels in images with impulse noise and a switching mechanism for smoothing the gradients of impulse noise samples and the gradients of impulse noise-free samples with different parameters. If the central pixel is impulse-like and has a high statistical value in gradient domain, the impulsive component of the weight in gradient bilateral filter is more heavily to suppress large impulses. It smoothes image toward a sharply-bounded, gradient piecewise-linear approximation which provides stronger noise reduction and better edge-limited smoothing behavior. Compared to most other spatial domain nonlinear filters, the proposed algorithm consistently yields good results in removing the mix of Gaussian and impulse noise and more notable edge-limited smoothing behavior. Like the trilateral filter, the proposed algorithm easily extends to N-dimensional signals.