Research On No-reference Image Quality Objective Assessment

Image quality objective evaluation is one of hottest research issue in the area of image processing,an effective image quality objective assessment method has great effects on optimizing the performance of image/video system and adjusting the parameters of image/video system.In this paper,we research plane image quality objective evaluation,an original no-reference image quality objective evaluation approach is raised to predict perceived quality of multiply-distorted plane images.Firstly,the images are transformed into local Walsh Hadamard transform maps by local Walsh Hadamard transform.Secondly,the features(i.e.,rotation invariant local binary pattern statistical features of zero sequency term and non-zero sequency terms)are extracted on local Walsh Hadamard transform maps.Finally,the extracted features and image quality scores are trained using support vector regression to form the plane image quality evaluation model,which implements mapping from the characteristic space to the image quality scores.Comprehensive evaluations are conducted on two plane image databases(i.e.,MLIVE database and MDID2013 database)and the result of the test shows the raised algorithm consists well with human subjective perception.Additionally,the performance of method is observably outperform the existence of better full-reference and no-reference plane image quality objective estimation methods.In this paper,stereoscopic image quality objective assessment is studied,a neoteric no-reference image quality objective evaluation approach is put forward to predict perceived quality of stereoscopic images.First,cyclopean images is synthesized using left and right views based on binocular rivalry characteristic.Then,cyclopean images are transformed into gradient magnitude maps and Laplacian of Gaussian maps respectively by calculating gradient and Laplacian of Gaussian,the features(i.e.,the statistical features of marginal probability distribution and dependency distribution)are respectively extracted on gradient magnitude maps and Laplacian of Gaussian maps.Finally,the extracted features are trained using support vector regression to form the stereoscopic image quality evaluation model,which implements mapping from the characteristic space to the image quality scores.The experimentations are performed on two stereoscopic image databases(i.e.,LIVE-3D database I and LIVE-3D database II)and the experimental results shows that quality objective prediction of the raised algorithm is effective,consisting well with human subjective perception.Besides,the performance of method has an advantage over the existence of better full-reference and no-reference stereoscopic image quality objective evaluation approaches.