Despite the vast amount research on visual search, we are still far from understanding such statements as "I cant find my keys because of all this clutter on my desk". The gap between research and real-life search exists in part because the stimuli used in research bear little resemblance to the reality of our surrounding visual world. To this end, we have been studying the impact of real-world clutter on object recognition and search.

Most recently, we have developed a scale invariant measure of clutter for natural images. This measure employs a segmentation algorithm at multiple scales to partition an image into "uniform" regions (shown on the right is an image and its segmentation at one scale). We have shown that the number of segmented regions across scale correlates well with human search times.

(Collaborative work with Mary Bravo)