No Reference Image Quality Assessment Using Block Based Features and Artificial Neural Network

In last decade, there has been explosive growth in multimedia technologies and its applications. For fast transmission, compression of image data is necessary. Due to this images are lead to distortion like blocking, ringing and blurring. The channel noise also gets introduced if transmitted over communication channel. Due to the distortion, image quality assessment plays an important role. In majority applications, original image is not available for reference. In such application, the metric which evaluates quality without reference is called "no reference quality" metric. Since human perception has limitation and in automated quality assessment application there is an immense need of developing no reference quality assessment framework. In this paper, we propose no reference image quality assessment scheme using the machine learning approach. Based on the degradation such as blocking, ringing artifacts, the related features such as average absolute difference between in-block image sample and zero-crossing rate, spatial frequency measure and spatial activity measures are computed for JPEG gray scale images. The earlier related work uses such parameters and mathematical predictors. Many time the correlation of extracted features, DMOS and output of predictor do not present correct assessment. In the proposed approach, properly trained back propagation artificial neural network with MOS as target is used. The result indicates that accuracy of quality assessment is better.