Purchase this article with an account.
Wietske Zuiderbaan, Jonathan van Leeuwen, Serge Dumoulin; Change detection: the role of low-level versus high-level image representations . Journal of Vision 2016;16(12):335. doi: 10.1167/16.12.335.
Download citation file:
© 2017 Association for Research in Vision and Ophthalmology.
Introduction Our internal representation of an image is not as detailed as we intuitively experience. This is exemplified by the fact that subjects fail to detect large changes in a visual scene, i.e. change-blindness. Current theories propose that the internal representation captures the gist (interpretation of the image) and that changes in gist are detected faster. On the other hand, we know that early visual cortex represents contrast energy. Here we investigate the role of the low-level feature contrast and the higher-level feature gist on our internal representation. Methods We measured reaction times (RTs) in a flicker-task using the change-blindness paradigm (Rensink,1997). We alternated two images (108 image-sets) and the subjects (n=60) indicated when and where they saw the change. The images were taken from the Berkeley Segmentation Dataset and Benchmark database (Martin et al,2001). This dataset contains manual segmentations where subjects identified the most important aspects of the image. We use these segmentations as a measure of gist. For every image-set, we computed both the change in local RMS-contrast and the change in gist. From these changes we defined 4 conditions: image-sets that are 'low' and 'high' in their differences for contrast and gist, respectively. We controlled for size, eccentricity, local contrast and luminance of the changed area. Results RTs were faster when image-sets were high in gist change (median increase RT = 2.2sec), or high in contrast change (median increase RT = 1.75sec). Furthermore, RTs were fastest when image-sets were both high in gist change and high in contrast change (median increase RT = 5.0sec). Discussion Our results suggest that the internal representation of the image, as measured with a change-detection paradigm, is not only influenced by high-level image interpretation (gist), but also by low-level image statistics such as contrast.
Meeting abstract presented at VSS 2016
This PDF is available to Subscribers Only