Color image fidelity metrics evaluated using image distortion maps

摘要: Abstract Several color image fidelity metrics are evaluated by comparing the metric predictions to empirical measurements. Subjects examined pairs consisting of an original and a reproduction. They marked locations on reproduction that differed detectably from original. We refer distribution error marks subjects as distortion maps . The empirically obtained compared predicited visible difference calculated using (1) widely used root mean square (point-by-point RMS) computed in uncalibrated RGB values, (2) point-by-point CIELAB Δ E 94 values (CIE, 1994), (3) S-CIELAB , spatial extension metric. RMS did not predict perceptual data well. provided better predictions, metric, which incorporated sensitivity eye, gave most accurate predictions. None excellent fit data. Image areas with poor were concentrated regions containing large negative local contrast. When these excluded our analysis, both had much agreement This suggests next step improving is redefine formula such terms