Full-Text Search:
Home|Journal Papers|About CNKI|User Service|FAQ|Contact Us|中文
《Journal of Optoelectronics·Laser》 2018-06
Add to Favorite Get Latest Update

A fast noise level estimation algorithm based on statistics feature value of wavelet transformed coefficients and depth neural network

XU Shao-ping;LIN Guan-xi;ZENG Xiao-xia;JIANG Yin-nan;TANG Yi-ling;School of Information Engineering,Nanchang University;  
Considering the fact that the noise can lead to regular change of the statistical feature values of the wavelet transformed coefficients and the powerful function approximation(mapping)capability of the deep neural network(DNN),a fast noise level estimation(FNLE)algorithm is proposed in this paper.Firstly the Daubechies 9/7 wavelet basis was utilized to transform the noisy image in three scales and three orientations,and then the generalized Gaussian distribution(GGD)model was used to model every subband coefficients.The two parameters of the GGD model were used as the feature values to describe the noise level,thus totally 18 feature values can be obtained to form a feature vector for describing the noise level of an image.Then,a large number of images corrupted with known noise levels were selected,their feature vectors were extracted,and the extracted feature vectors and corresponding noise levels were used to construct training data set.Finally,the DNN technology was employed to train a prediction model on the training data set,which can map agiven feature vector to the corresponding noise level.In order to obtain more accurate prediction,the range of noise level was subdivided into five sub-ranges,and more accurate prediction models were trained.Compared with the state-of-the-art algorithms,the FNLE algorithm adopts a quite different training-based approach.Once the prediction model is obtained,its execution time is very short.At the same time,DNN technology ensures the accuracy of the estimation.Extensive experimental results verify that the proposed FNLE algorithm has better advantages in terms of both the estimation accuracy and the efficiency.
【Fund】: 国家自然科学基金(61662044 61163023 51765042 81501560);; 江西省自然科学基金(20171BAB202017)资助项目
【CateGory Index】: TP183;TP391.41
Download(CAJ format) Download(PDF format)
CAJViewer7.0 supports all the CNKI file formats; AdobeReader only supports the PDF format.
©2006 Tsinghua Tongfang Knowledge Network Technology Co., Ltd.(Beijing)(TTKN) All rights reserved