2D And 3D Image Quality Assessment Algorithm Research Based On Contourlet Transform

The images can reproduce the real scene in the mind of people.Stereo images are more vivid and can give a new visual enjoyment.But in the process of production,storage,and transmission of the images,the introductions tend to distortion.The assessment of good points and the bad points about the quality of 2D and 3D images has become an important issue.Image quality evaluation mainly includes two kinds: subjective quality evaluation and objective quality evaluation.Subjective quality evaluation is a method of evaluating the score directly by the human eye.But the objective quality evaluation is not based on the human eye scoring,It's evaluate the image quality automatically through computers.Objective quality evaluation method has better practicability and generalization.Contourlet transformation has the characteristics of multi-resolution,multi-scale,multi-direction and anisotropy.These characteristics are similar to the human eye.Based on this,in order to obtain an objective image quality evaluation method which is more accurate and more consistent with human eye characteristics,this paper carries out the quality evaluation algorithm of 2D and 3D image based on the Contourlet transform.The main research contents are as follows:(1)Put forward the quality evaluation method of no-reference 2D image which is based on non sampling Contourlet transform.First the method transforms 2D image into subbands through Nonsubsampled Contourle,and then obtains transform subbands and extract high frequency energy value,average gradient,structural similarity index,spatial correlation index,gradient direction features.Finally these features are input to the supported vector regression(SVR)and can be obtained the evaluation scores through learning.(2)Put forward the quality evaluation method of no-reference 3D images which is based on disparity information and Contourlet transform.This method is mainly using SSIM stereo matching algorithm to extract the disparity map and matching difference map,with the left and right image,using Contourlet transform to get transform subband,and then exact the high frequency subband transform,energy index,edge strength,information entropy feature,using supporting vector regression learning to get evaluation scores.(3)The leading edge research of image quality evaluation is mainly aiming at the application of the hottest depth learning algorithm in the field of machine learning and image processing in image quality assessment.The decomposition of the image using Contoulet transform,then the low frequency coefficients of the reconstructed image is zero,using convolutional neural network extract features and regression analysis training on the training set get the Model,using Model to predict the test set,and finally get the evaluation scores.