Classification of image distortions using image quality metrics and linear discriminant analysis

Numerous Image Quality Measures (IQMs) have been proposed in the literature. While some are based on structural analysis of images, others rely on the cha-racteristics (or limitations) of the Human Visual Sys-tem (HVS). However, none of the existing IQMs is shown to be robust across all types of degradations. Indeed, some IQMs are more efficient for a given arti-fact (such as blurring or blocking) but inefficient for others. In this paper, we propose to circumvent this limitation by adding a preprocessing step before mea-suring image quality. We propose to detect the type of the degradation contained in the image, then use the most "relevant" IQM for that specific type of degrada-tion. The classification of different degradations is performed using simple Linear Discriminant Analysis (LDA). The performance of the proposed method is evaluated in terms of classification accuracy across different types of degradations and shown to outper-form different IQMs when used independently of the degradation type.