Change Blindness Is Influenced by Both Contrast Energy and Subjective Importance within Local Regions of the Image

Abstract

Our visual system receives an enormous amount of information, but not all information is retained. 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 our ability to detect these changes is influenced by the gist or interpretation of an image. On the other hand, stimulus-driven image features such as contrast energy dominate the representation in early visual cortex (De Valois and De Valois, 1988; Boynton et al., 1999; Olman et al., 2004; Mante and Carandini, 2005; Dumoulin et al., 2008). Here we investigated whether contrast energy contributes to our ability to detect changes within a visual scene. We compared the ability to detect changes in contrast energy together with changes to a measure of the interpretation of an image. We used subjective important aspects of the image as a measure of the interpretation of an image. We measured reaction times while manipulating the contrast energy and subjective important properties using the change blindness paradigm. Our results suggest that our ability to detect changes in a visual scene is not only influenced by the subjective importance, but also by contrast energy. Also, we find that contrast energy and subjective importance interact. We speculate that contrast energy and subjective important properties are not independently represented in the visual system. Thus, our results suggest that the information that is retained of a visual scene is both influenced by stimulus-driven information as well as the interpretation of a scene.

Keywords: change detection; contrast energy; natural images; scene perception; subjective importance.

FIGURE 1

Five example images (A) from the ‘Berkeley Segmentation Dataset and Benchmark’ database (Martin et al., 2001). For all the images of this database human observers manually identified the important aspects of the image (B). Different observers drew the labels, and their task was to draw lines on the image to highlight the parts of the image they considered to be important for the representation of the scene. We used the average manually labeled images of five observers as our definition of the subjective importance of the image. The pixels of the manually labeled images have values between 0 (not labeled) and 1 (pixel labeled by all 5 observers).

FIGURE 2

Examples of image pairs that were used in the experiment. The original images were taken from the ‘Berkeley Segmentation Dataset and Benchmark’ database (Martin et al., 2001). The arrows indicate where the manipulation in the image was made (the arrows were not present in the actual experiment). For every condition we show one example image pair (A–D). The conditions are based on the amount of change in contrast energy and subjective importance. The conditions were balanced for changes in size, distance from the center, contrast, luminance and spatial frequency.

FIGURE 3

An illustration of the procedure of calculating the change in contrast energy and subjective importance. Two example images [taken from the ‘Berkeley Segmentation Dataset and Benchmark’ database (Martin et al., 2001)] are shown with changes predominant in subjective importance (A) and contrast energy (B). The red line represents the region manipulated to alter the image. From this region we calculated the difference in local contrast energy as well as the amount of subjective importance.

FIGURE 4

The distribution of local contrast energy change (A) and change in subjective importance (B) for all the image pairs. The vertical striped black line indicates the 50th percentile used to define our four different conditions. Image pairs left of the 50th percentile were used in the ‘low’ condition, both for contrast energy change and subjective importance change. Image pairs right to the 50th percentile were used in the ‘high’ conditions.

FIGURE 5

Proportion of detections (hits), detections in which subjects failed to indicate the correct location (false alarms) and the failure to detect the change (misses) for the conditions based on contrast energy (CE) and subjective importance (SI). The averaged data are the mean from all subjects, and the error bars reflect the 95% confidence interval. No significant differences were found.

FIGURE 6

(A) The median reaction times of all the correct detections for all subjects for the different image conditions. The striped line represents the median RT of the race-model. The error bars reflect the bootstrapped 95% confidence interval. (B) The cumulative distributions of the correct responses for the different conditions. We analyzed our results using a GLM approach, with the inverse Gaussian as a linking function. The striped lines are the fits to the data with an inverse Gaussian function. Note that we only show the RTs up to the first 40 s of the experiment here, the entire RT-range is 0–240 s. We found significantly shorter reaction times for both changes in contrast energy (CE) and subjective importance (SI). Furthermore, we found a significant interaction effect, i.e., reaction times were shorter when changes affected both contrast energy and subjective importance. (C) The comparison of the CDF of the condition high-contrast/high-subjective importance with the CDF of the race-model. The CDF of the race-model is the summed CDFs of the conditions high-contrast/low-subjective importance and low-contrast/high-subjective importance. We found statistical significant shorter RTs for the condition high-contrast/high-subjective importance compared to the race-model for the 10th to the 70th percentile. This indicates that the interaction effect cannot be explained by statistical facilitation alone.