查看文章

• Census: Census converts each image patch into a binary descriptor by checking for each pixel whether it is lighter or darker than the patch center pixel [4]. Two patches’ difference is computed as their descriptors’ Hamming distance.

• iCID: The improved color image difference metric [2] predicts the perceived difference between two color images by normalizing them to reference viewing conditions (accounting for the viewing distance) and transforming the resulting images into a perceptually uniform color space without perceived cross-contamination of color attribute predictors (lightness, chroma and hue). Within the color attribute channels, local-average, local-contrast and local-structure-based difference features are extracted and merged to a single value reflecting the perceived difference between the images. We analyzed six variants (iCID perceptual, huepreserving, saturating, perceptual csf, huepreserving …