Focused pooling for objective quality estimation

We present an investigation into a simple method for focused pooling, the process of aggregating local signal quality estimates into a global quality score in objective quality metrics. Applicable to a wide array of signal quality tasks, we show that the proposed method can improve performance of a number of established objective metrics on several extensive subjectively annotated datasets in two application domains: image quality and image fusion performance. Defining an effective regional aggregation strategy the method provides a simple model for the relatively well known notion that human observers base their opinion on a section of the perceived signal only when deciding on visual quality. Our results support the theory that it is in fact the worst section of any visual signal that determines its perceptual quality.