Features for Predicting Quality of Images Captured by Digital Cameras

Algorithmic image quality metrics have been based on the assumption that an image is only distorted by a single distortion type at a time. The performance of the current metrics is low if image concurrently includes more than one distortion. The aim of this study was to find efficient feature sets for predicting visual quality of real photographs which are subjected to many different distortion sources and types. Features should support each other and function with many concurrent image distortions. We used correlation based feature selector method and image database created with various digital cameras for feature selection. Based on the study the results are promising. Our general and scene-specific feature combinations correlate well with the human observations compared to the state-of-the-art metrics.