DCT-block Difference Based Noise Estimation Using An Iterative Framework

Blind noise level estimation is an important step of many image processing applications,such as denoising,compression,and segmentation.In this paper,an iterative noise estimation algorithm based on DCT of image patches is proposed.The proposed method is mainly enlightened from the aspects of frequency domain and the weak texture block.The image is a two-dimensional matrix in the spatial domain,each element of the matrix corresponds to a pixel of the image.Spatial domain of an image delivers direct visual information to people,but it's transformed domain reflects change of gray level.And in the transformed domain,we can separate the noise and image information more easily.Therefore,this paper selected DCT-based method for image noise estimation.Smooth area is robust for the selection of weak texture blocks,and appears a good adaptability to different types of images,thus we would select weak texture image blocks from the smooth area for noise estimation.The main steps of the proposed algorithm are as follows:(1)Firstly,processed the noisy image into DCT domain,and divided it into several image blocks that are of one size and non-overlap for each other,then process these blocks by an adaptive denoising method.Next,this paper proposes a new weak texture image block detector,which would identify weak texture image blocks through the ratio between DC and AC coefficients.(2)Secondly,forming a three-dimensional matching block group through the similarity of the weak texture image blocks.This paper proposes a new noise estimation model,obtain the matrix of DCT-based difference from each group,and collect all of them to form a big data set that obey normal distribution.Then we can get the related parameters of noise level through the study of standard deviation of the dataset(3)Finally,we reference the ideas of iterative and design a framework in order to estimate the noise level more stable and precise.Algorithm would run unless the result tend to be stable or it has cycled for several times.The latter is to ensure the speed of proposed algorithm.In general,Our noise estimation method is fast and effective among the algorithms that based on DCT?We have demonstrated the good performance of this method using several public image databases,and the method get a considerable effect on the estimation of low level noise,which is difficult to estimate for the noise estimation field.