Image Haziness Contrast Metric Describing Optical Scattering Depth

Abstract

Contrast is not uniquely defined in the literature. There is a need for a contrast measure that scales linearly and monotonically with the optical scattering depth of a translucent scattering layer that covers an object. Here, we address this issue by proposing an image contrast metric, which we call the Haziness contrast metric. In its essence, the Haziness contrast compares normalized histograms of multiple blocks of the image, a pair at a time. Subsequently, we test several prominent contrast metrics in the literature, as well as the new one, by using milk as a scattering medium in front of an object to simulate a decline in image contrast. Compared to other contrast metrics, the Haziness contrast metric is monotonic and close to linear for increasing density of the scattering material, compared with other metrics in the literature. The Haziness contrast has a wider dynamic range, and it correctly predicts the order of scattering depth for all the channels in the RGB image. Utilization of the metric to evaluate the performance assessment of dehazing algorithms is also suggested.