An adaptive bilateral filter based framework for image denoising

Conventional bilateral filter (BF) can suppress Gaussian noise effectively, but fail to remove impulsive noise and may blur edges in an image. To address these shortcomings, we aim to develop an improved bilateral filter based framework which is capable of effectively removing universal noise, i.e. impulses, Gaussian noise or mixture of the two types of noises, from images without oversmoothing edge details. To this end, our proposed denoising framework mainly consists of an impulse noise detector (IND), an edge connection precedure and an adaptive bilateral filter (ABF). Specifically, we first compute an edge component value to classify a pixel into impulse or nonimpulse. This is followed by an edge connection procedure, producing more connected edge regions. Then we introduce an adaptive bilateral filter which switches between Gaussian and impulse noise depending on the impulse noise detection results. This makes the adaptive bilateral filter be robust to these two types of noises. We also present an improved artificial bee colony (IABC) algorithm to optimize the parameters of the adaptive bilateral filter, enabling both effective noise removal and fine edge preservation. Experimental results demonstrate that the proposed image denoising framework outperforms alternative state of the art filters both in visual qualitative evaluations and quantitative comparisons.