A new image quality estimation approach for JPEG2000 compressed images

No Reference (NR) Image Quality Metrics (IQMs) for JPEG2000 compressed images consider generally the ringing effect as the dominant degradation. However, for a certain bite rate, blur degradation appears and becomes the most annoying. We propose here to estimate the quality of JPEG2000 compressed images by first identifying the dominant distortion and then selecting the appropriate metric (blur or ringing). The degradation identification is here realized using an Artificial Neural Networks (ANN). The performances of the proposed method are evaluated in terms of classification accuracy and correlation with subjective scores.