Evaluation of a multiscale color model for visual difference prediction

Abstract

How different are two images when viewed by a human observer? There is a class of computational models which attempt to predict perceived differences between subtly different images. These are derived from theoretical considerations of human vision and are mostly validated from psychophysical experiments on stimuli, such as sinusoidal gratings. We are developing a model of visual difference prediction, based on multiscale analysis of local contrast, to be tested with psychophysical discrimination experiments on natural-scene stimuli. Here, we extend our model to account for differences in the chromatic domain by modeling differences in the luminance domain and in two opponent chromatic domains. We describe psychophysical measurements of objective (discrimination thresholds) and subjective (magnitude estimations) perceptual differences between visual stimuli derived from colored photographs of natural scenes. We use one set of psychophysical data to determine the best parameters for the model and then determine the extent to which the model generalizes to other experimental data. In particular, we show that the cues from different spatial scales and from the separate luminance and chromatic channels contribute roughly equally to discrimination and that these several cues are combined in a relatively straightforward manner. In general, the model provides good predictions of both threshold and suprathreshold image differences arising from a wide variety of geometrical and optical manipulations. This implies that models of this class can be generally useful in specifying how different two similar images will look to human observers.

A general rule for sensory cue summation: evidence from photographic, musical, phonetic and cross-modal stimuli

“…The same Minkowski rule (typically with m ¼ 3-4) has been used to model the detection of changes in natural visual images, when multiple tiny cues are contributed across very many visual channels or model neurons [16 -18]. Such modelling has been extended to the perceived magnitudes of suprathreshold differences in natural images similar to those that we have described here [18,25]. We have shown here that the same sort of Minkowski exponent describes the perception of suprathreshold changes in naturalistic auditory stimuli, as well as visual changes.…”

Search for gross illumination discrepancies in images of natural objects

“…In order to manipulate the heterogeneity among the pebbles present within a stimulus image, the similarity of each pebble was estimated using a visual difference predicting (VDP) model (Lovell, Párraga, Ripamonti, Troscianko, & Tolhurst, 2006;. For each pebble, one of the four available orientations was randomly chosen.…”

Analyzing Visual Signals as Visual Scenes

“…Though information on human form perception is sparse compared to color information, future methodological developments could try to begin incorporating species differences in form perception, just as is now happening for interspecific color perception. Some techniques which are promising avenues for progress towards this goal include use of biologically inspired segmentation algorithms [Lindbald and Kinser, 2005] and models of cortical processing of spatiotemporal information [Lovell et al, 2006].…”