RelContrast

RelContrast is a metric used in imaging and vision to quantify the relative contrast between two regions, objects, or features within a scene. The concept focuses on how strongly one region stands out from another, normalized by the overall brightness to enable comparisons across scenes with different illumination.

To compute it, luminance is typically estimated from the image, often using the Y channel of a luminance-chrominance color space such as YUV or YCbCr, or the V channel of HSV. If two regions A and B have mean luminances LA and LB, a common definition is RelContrast = |LA − LB| / (LA + LB). This yields a value between 0 and 1, with higher values indicating greater relative contrast. Variants may use median, peak, or region-distribution-based statistics, producing slightly different results.

RelContrast has applications in image quality assessment, tone mapping and high dynamic range imaging, adaptive exposure control, and region-based feature detection or segmentation. Its appeal lies in being relatively scale-invariant to uniform lighting changes, yet it remains sensitive to how regions are chosen, to noise, and to the specific method used for extracting luminance or chose color space.

It is not a single universal standard; multiple formulations exist in the literature, and practitioners may adapt the metric to their dataset and task. In practice, RelContrast is often used alongside other metrics to characterize perceptual contrast and to guide processing pipelines in imaging and computer vision.