Feature Prioritization in Classification of Novel Objects

novel shapes designed for the experiment

Object classification is essential to human learning as it helps us cope with various stimulus around the world. Regardless of multiple features within a single object, object classification seems to occur seamlessly within our cognitive process. In this experiment, we test how we prioritize each feature within an object and how these features are weighted when we categorize a certain object. Test subjects were given novel shapes that each featured either size, color, or orientation, and had to determine whether the shape belongs to a category of a given prototypical shape.

The preliminary result showed that color was the single most determining feature when categorizing an object, showing 72.6% of incorporation in all trials, while orientation was the least with 60.7%, but the differences were not statistically significant. We further went on to use logistic regression to analyze the result, which showed thresholds for identifying a novel object to be in a certain category. However, these thresholds for each feature was not significantly different. The experiment suggests that categorization is more of an elaborate and holistic process that combines different features when categorizing a novel object.

Avatar
Jin Jeon
UX Researcher

Related