Inherent limitations of Visual Search and the Role of InnerScene Similarity.pdfVIP

Inherent limitations of Visual Search and the Role of Inner-Scene Similarity Tamar Avraham and Michael Lindenbaum Computer Science Department, Technion Haifa, Israel 32000 tammya, mic@cs.technion.ac.il Abstract [21], aiming to explain the variability of search behavior, suggest that some visual properties (such as color, orienta- This work focuses on inner-scene objects similarity as an tion, spatial frequency, movement) are extracted in parallel information source for directing attention and for speeding- across the visual field. Under this hypothesis, if a target up visual search. A scalar measure (similar to Kol- is unique in such a feature, it pops-out and gets immediate mogorov’s -covering of metric spaces) is suggested for attentional priority. Otherwise, it is detected only after a quantifying how much a visual search task can benefit from longer serial search. While several aspects of the Feature this source of information. The measure provided is algo- Integration Theory were criticized, the theory was domi- rithm independent, providing an inherent measure for tasks’ nant in visual search research and much work was carried difficulty, and can be also used as a predictor for search out, e.g. to understand how feature integration occurs (some performance. We show that this measure is a lower bound examples are [10],[25],[23]). Duncan and Humphreys re- on all search algorithms performance and provide a simple jected the dichotomy to parallel vs. seria