Item

Abstract

The human visual system is able to extract a vast amount of scene information from a briefly shown picture. Here, we investigated the ability to accurately and quickly estimate horizon position, which is related to viewer orientation and scene structure in general. In the first, perceptual study, we asked 18 participants to estimate the horizon position after a 150 ms, masked presentation of typical outdoor scenes from different scene categories. All images were shown in upright, blurred, inverted, and cropped conditions to investigate the influence of different information types on the perceptual decision. Overall, participants were fairly consistent in their estimates (r=0.62); inverted images, however, produced significantly worse estimates, indicating the importance of global scene consistency. In the second, computational experiment, we correlated the performance of several algorithms for horizon estimation with the human data—algorithms ranged from simple estimations of bright-dark-transitions to more
sophisticated frequency spectrum analyses motivated by previous computational modeling of scene classification results. Correlations between human and computational estimates ranged from 0.2 to 0.6 and varied by algorithm and scene category. Overall, frequency spectrum analysis provided the best results, which taken together with the perceptual data, highlights the importance of global, frequency-based information in scene processing.