Information extraction from shadowed regions in images: an eye movement study

Visual attention in change blindness for objects and shadows

Studies have found that observers pay less attention to cast shadows in images than to better illuminated regions. In line with such observations, a recent study has suggested stronger change blindness for shadows than for objects (Ehinger et al., 2016). We here examine the role of (overt) visual attention in these findings by recording participants' eye movements. Participants first viewed all original images (without changes). They then performed a change detection task on a subset of the images with changes in objects or shadows. During both tasks, their eye movements were recorded. In line with the original study, objects (subject to change in the change detection task) were fixated more often than shadows. In contrast to the previous study, better change detection was found for shadows than for objects. The improved change detection for shadows may be explained by the balancing of trials with object and shadow changes in the present study. Eye movements during change detection indicated that participants searched the bottom half of the images. Shadows were more often present in this region, which may explain why they were easier to find.

Evaluating refactorings for disciplining #ifdef annotations: An eye tracking study with novices

The C preprocessor is widely used in practice. Conditional compilation with #ifdef annotations allows developers to flexibly introduce variability in their programs. Developers can use disciplined annotations, entirely enclosing full statements with preprocessor directives, or undisciplined ones, enclosing only parts of the statements. Despite some debate, there is no consensus on whether a developer should use exclusively disciplined annotations. While one prior study found undisciplined annotations more time-consuming and error-prone, another study found no difference between disciplined and undisciplined annotations regarding task completion time and accuracy. In this article, we evaluate whether three fine-grained refactorings to discipline #ifdef annotations correlate with improvements in code comprehension and visual effort with an eye tracker. We conduct a controlled experiment with 64 human subjects who were majoritarily novices in the C programming language. We observed statistically significant differences for two refactorings to discipline annotations with respect to the analyzed metrics (time, fixation duration, fixation count, and regressions count) in the code regions changed by each refactoring.