Parametric modeling of visual search efficiency in real scenes.

Xing Zhang,Baoding Zhou,Qingquan Li,Zhixiang Fang,Qin Zou,Suliann Ben Hamed

doi:10.1371/journal.pone.0128545

Abstract

How should the efficiency of searching for real objects in real scenes be measured? Traditionally, when searching for artificial targets, e.g., letters or rectangles, among distractors, efficiency is measured by a reaction time (RT) × Set Size function. However, it is not clear whether the set size of real scenes is as effective a parameter for measuring search efficiency as the set size of artificial scenes. The present study investigated search efficiency in real scenes based on a combination of low-level features, e.g., visible size and target-flanker separation factors, and high-level features, e.g., category effect and target template. Visible size refers to the pixel number of visible parts of an object in a scene, whereas separation is defined as the sum of the flank distances from a target to the nearest distractors. During the experiment, observers searched for targets in various urban scenes, using pictures as the target templates. The results indicated that the effect of the set size in real scenes decreased according to the variances of other factors, e.g., visible size and separation. Increasing visible size and separation factors increased search efficiency. Based on these results, an RT × Visible Size × Separation function was proposed. These results suggest that the proposed function is a practicable predictor of search efficiency in real scenes.

Highlights

One important skill is the ability to search for and visually identify target objects among irrelevant local distractions in real-world scenes
We focused on three questions in the present study: (1) the quantitative contribution of multiple factors to search efficiency in real scenes; (2) whether Ds is a suitable descriptor of crowding in real scenes; (3) whether the reaction time (RT) for real objects can be quantitatively measured by a combination of low-level and high-level features
We proposed a computational model of search efficiency in real scenes

Summary

Introduction

One important skill is the ability to search for and visually identify target objects among irrelevant local distractions in real-world scenes (for example, searching for a hotel in the street, a book on bookshelves, or a vehicle in a parking lot). Owing to the complexity of real scenes and limited neural resources, visual search efficiency relies on a selection mechanism known as visual attention [1,2], which enables humans to allocate more neural resources to extracting the most important information from the physical environment. In which observers intend to search for a pre-defined target among irrelevant distractors, is one of the most important paradigms for studying visual attention [3,4]. Extensive studies on visual search have greatly improved our understanding of the mechanisms of attention deployment.

Methods

Results

Conclusion