Saccadic Target Selection

Introduction: When the visual system selects a peripheral target as its next region of interest, it has not only computed the target’s spatial location, but it has already selected a preferred landing point within the selected target structure! This had already been shown in experiments made 40 years ago by Richards and Kaufman:

But nobody has really investigated whether this also takes place during free-viewing of natural scenes.

Method: We used Tatler’s data and my implementation of the symmetric-axis transform to relate fixation points to their nearest symmetric axis.

Results: We found that saccades toward L features and parallel lines landed in specific locations, similar to what has been already measured by Richards and Kaufman [manuscript as html].

Conclusion: Even during free-viewing of natural scenes, the visual system selects preferred locations within a structure. No neural network or computer vision algorithm can make these predictions. Bottom-line: it requires a novel model of the translation-independent structure that allows computing the preferred target points. This is exactly what I am working on, see my decomposition.

[manuscript as html, pdf] describes our study

[link to sym-axis] describes the symmetric-axes transform

[link to funding website] the website of the 5-lab European collaboration