The contrast statistics of natural images can be adequately characterized by a two-parameter Weibull distribution. Here we show how distinct regimes of this Weibull distribution lead to various classes of visual content. These regimes can be determined using model selection techniques from information theory. We experimentally explore the occurrence of the content classes, as related to the global statistics, local statistics, and to human attended regions. As such, we explicitly link local image statistics and visual content.

Can a machine learn to perceive emotions as evoked by an artwork? Here we propose an emotion categorization system, trained by ground truth from psychology studies. The training data contains emotional valences scored by human subjects on the International Affective Picture System (IAPS), a standard emotion evoking image set in psychology. Our approach is based on the assessment of local image statistics which are learned per emotional category using support vector machines. We show results for our system on the IAPS dataset, and for a collection of masterpieces. Although the results are preliminary, they demonstrate the potential of machines to elicit realistic emotions when considering masterpieces.

The selection of salient regions in an image is the first step in many computer vision algorithms, e.g. objects recognition, classification or tracking. Our hypothesis is that local image statistics are indicative for saliency. We combine natural image statistics with the detection of salient regions. Particularly, we consider the integrated Weibull distribution as a parameterized model, which provides a good fit to the statistics of natural images. Here we show how distinct regimes of the integrated Weibull distribution leads to various local saliency mechanisms. With model selection techniques from information theory, we can determine the probability for every distinct regime, to explain the statistical properties of local image content. Hence, resulting in new algorithms for salient region detection.

Knowing why we direct our gaze to particular locations in images is important for understanding image interpretation. We hypothesized that under natural free viewing conditions our gaze is first drawn to image regions that statistically differ from the rest of the image. To test this, we computed local image statistics for the regions where subjects fixated during a free viewing task and compared these to the statistics of randomly selected (non-fixated) patches. In particular, we focus on the distribution of contrasts and edges in natural images, which is well described by the two-parameter Weibull distribution. Besides the contrast parameter, we also consider edge frequency as an additional attractor. In the experiment, we used National Geographic photos as stimuli. Our results demonstrate significantly different distributions of Weibull parameters for fixated and non-fixated image regions. The results identify both contrast and edge frequency to be cues for attention. Hence, natural image statistics as captured by the two-parameter Weibull distribution could play a role in determining where we direct our first few saccades.

Spatial statistics of natural images can be divided into three categories, each following different distributions as determined by statistical algorithms (Geusebroek and Smeulders, 2003 Ninth IEEE International Conference on Computer Vision volume 1, p. 130). Two of these classes (random and regular) are associated with local texture perception. We hypothesized that regular textures draw attention and wondered whether this classification can predict bottom - up saccade behaviour. While we tracked their gaze, subjects (N=5) made saccades to circularly arranged arrays of 8 textures (one of them classified as regular, the other random; textures were from the Columbia - Utrecht texture database: http://www.cs.columbia.edu/CAVE/curet/). In a total 3877 trials, 851 saccades went to the location with the regular texture, whereas each of the other 7 locations received on average 432 saccades. This indicates that the presence of a texture classified as regular nearly doubled the chance of subjects directing their gaze to that particular position. While these results provide no proof that humans classify the world into regular and random, they suggest that the brain may compute something analogous to this statistic to determine regions of interest.