1 Psychology and human development1.1 Sensation and perception
1.1.5 Feature networks and object recognition
1 min read
Written by Resident Doctor Committee · Published
New in the 2027 syllabus
The College added this topic for exams from January 2027.
From features to objects
Feature detectors respond selectively to simple properties, such as edges or orientation. Networks combine their responses to represent more complex patterns.
Feature networks link these levels of representation. Active features support candidate objects. Processing also runs downwards: what the context leads us to expect can influence lower-level processing.
Recognition by components
Biederman proposed recognition by components. In this model, an object is represented as an arrangement of simple three-dimensional parts called geons.
Geons include cylinder-like and cone-like parts. Objects made from similar components are told apart by how those parts are arranged.
Recognition with incomplete input
Redundancy means several features or cues can support the same interpretation. Recognition can therefore survive noise or an object being partly hidden (occlusion).
Diagnostic junctions, points where contours meet, can still show how the components fit together even when much of the outline is hidden.
Viewpoint tolerance is a strength of component accounts. Even so, unusual views and fine distinctions between similar objects can still cause difficulty.
Gestalt grouping explains how parts form perceptual wholes. Component models propose representations that support an object's identity and meaning.