It is need of the decade to develop systems that evaluate the quality of an image as it saves various subsequent unnecessary noise filtering applications in image processing. In this paper, we have presented a erence Image Quality Assessment (NRIQA) technique for prediction of image quality of JPEG compressed images. We are using computational training model which is trained with DWT based features. We have modified the frequency domain feature extraction scheme by using a novel mean coefficient DWT technique. We have tested the results on the LIVE database. © 2016 IEEE.