Mean shift point-mass level-of-detail

This paper proposes a novel image detail metric based on the mean shift segmentation procedure in which level-of-detail is expressed as a measure of local point concentration. Regions containing significant numbers of points are identified as low detail, while those made up of fewer points are assigned higher detail scores. This somewhat counterintuitive approach is based on the understanding that points in low detail and homogeneous regions are highly clustered, while those found in more detailed and diverse regions are not. Exact point concentrations vary considerably over a photograph, producing a multitude of intermediate detail levels. Local detail scores are computed for each segmented image region. Once combined, these values form a global detail map. This research is primarily motivated by a desire to quantify picture detail in the field of image quality analysis.