AOI Extraction

The aoi_extraction submodule has transformers to identify AOI areas for input fixations. As well as standalone transformers, there is a convenient meta transformer that allows to specify all desired methods and use several AOI extraction methods at the same time.

• ShapeBased
• ThresholdBased
• GradientBased
• OverlapClustering

AOI Matcher

There is also a class that works with already defined Areas of Interest. It can merge regions together if they are strongly intersected. Formally, it minimizes Shannon Entropy. It is strongly recommended to use it as post-processor after any of above transformers.

• AOIMatcher

Reference

The algorithms are taken from Wolfgang Fuhl. “Image-based extraction of eye features for robust eye tracking”. 2019.