====== Attentional saliency in images ======

[[teaching:methods|Back to teaching methods main page]]

\\
Students pick an image and circle regions that appear to draw attention to themselves: salient regions.  

The instructor runs the images through the Itti-Koch saliency algorithm, which marks pixels that ought to jump out at viewers because they represent regions that are different from their surroundings.

\\
[[teaching:saliency:saliency_instructions|Background and instructions given to students]]

\\
[[https://github.com/acate/saliency_maps|Github page for source code]]

\\
===== Sample results =====

\\
The algorithm identifies image regions that stand out based on several basic visual features, including color, intensity and edge orientation.

{{:teaching:saliency:person18_fig.png|}}

\\
It sometimes appears that the algorithm is picking out complex objects that are similar to what humans ultimately notice ... 


{{:teaching:saliency:person15_fig.png|}}

{{:teaching:saliency:person1_fig.png|}}


... however, the algorithm actually works in "local" fashion based on small, often meaningless visual differences.  

{{:teaching:saliency:person9_fig.png|}}

{{:teaching:saliency:person11_fig.png|}}