Image restoration using modified Anscombe transform and non linear multiresolution median transform

Image restoration is the important process for getting back the original image, without any loss of the information. The capturing processes of imaging devices are subjected to different signal dependent noise and they are considered as Poisson-Gaussian noise. The removal of such signal dependent noise is performed by applying nonlinear variance stabilization transformations (VST) such as modified Anscombe transform. This transform stabilizes the noisy data and makes the data to have Gaussian noise distribution with a constant variance. Here, a multiscale variance stabilization transform (MS-VST) based on modified Anscombe transform has been proposed which is extremely suitable for low intensity region and can be applied to any dimensional data. The Gaussian noise can be removed by using multiresolution median transform which consists of series of smoothing of the input image. The various operations performed in this transform are median filtering, decimation and interpolation filtering. The desired noise free image can be obtained by applying an exact unbiased inverse of Anscombe transformation. It completely removes the bias error which arises when the nonlinear forward variance stabilization transform is applied. The proposed MS-VST combined with exact unbiased inverse of Anscombe transform plays an integral part in providing accurate denoising result particularly when the input SNR is very low.