Single image de-hazing using adaptive dynamic stochastic resonance and wavelet-based fusion

Abstract

This paper presents enhancement-based single image de-hazing algorithms using dynamic stochastic resonance (DSR) and wavelet fusion. The initial algorithm employs DSR controlled with image perceptual quality metrics to automatically guide the process in an adaptive way via intersecting curves. The extent of de-hazing is determined by a parameter dynamically computed from the relevant regions of the segmented hazy image. The second algorithm combines the strengths of the DSR-based algorithm, partial differential equation (PDE), fractional calculus (FC) and multi-scale filtering-based de-hazing techniques via wavelet fusion. The problems addressed include colour distortion, image darkening, halo effects, edge, noise and sky region over-enhancement. Results indicate that the proposed approaches yield comparable or better results than several de-hazing algorithms from the literature in terms of visual clarity, objective metrics and runtime.