Characterizing the perceived quality degradation of still-camera motion blur

TR Number



Journal Title

Journal ISSN

Volume Title


Virginia Tech


Camera motion blur was examined in two experiments, in an attempt to behaviorally validate candidate predictors of the associated degradation in photographic image quality.

In the first experiment, motion of a hand-held 35-mm single-lens-reflex camera was measured using inertial sensors while film was simultaneously exposed. Thirty-two subjects photographed two studio scenes at 1/60-s, 1/30-s, 1/15-s, and 1/8-s shutter speeds, replicated four times. Several spatial image quality metrics (blur size, Subjective Quality Factor, SQF; Modulation Transfer Function ,Area, MTFA; and Equivalent Passband, Ne), derived analytically from the motion data, were used as dependent measures for statistical analysis.

All three modulation-transfer-function (MTF)-based analytic quality metrics were strongly intercorrelated. Under these measures, the 1/60-s shutter speed yielded significantly higher quality photographs than the other speeds. The 1/8-s condition proved the worst. A difference between the 1/30-s and 1/15-s conditions was not reliably supported under the MTF-based metrics, although a significantly greater blur size resulted on average in the longer exposure. Tripod-mounted exposures compared favorably with the hand-held shots at 1/15 s and 1/8 s, but differences were not statistically supported at the two faster speeds.

In the second experiment, 16 prints of each scene were selected from those collected in the first experiment. The log-transformed analytic SQF values were used to select the prints, ranging from worst to best at equal increments. Measures of perceived image quality and sharpness were derived from psychophysical judgments by 48 observers.

Linear regression analyses identified several analytic models that predicted substantial amounts of the variance in mean judgments across all observers. The analytic SQF, MTFA, and Ne metrics based on the products of the unidimensional values in x and y each accounted for roughly 90% of the variance in relative judgments. The use of any of these measures is recommended for the evaluation of motion blur in future camera-shake studies. As a secondary choice, for computational ease, the blur size can be used.