Perceptual Geometry

Title: Perceptual Geometry

Speaker: Lakshman Prasad

Abstract:

The act of perception as a pre-attentive imposition of order on the complexity of visual stimuli leading to understanding of a scene, an image, or a drawing has fascinated scientists long before the experiments of Max Wertheimer in his Leipzig laboratory.

Today, a growing number of researchers in computer vision are adopting insights from empirical findings on the workings of perception to design algorithms that extract visually meaningful features from images.

In this spirit, this talk briefly introduces a geometric framework for modeling elementary perceptual organization notions to achieve tasks such as image segmentation and shape decomposition efficiently using low-level visual cues such as edges and their regional relationships. The framework employs Delaunay triangulation as an image adaptive grid that relates proximate edge elements. Image segmentation and shape decomposition tasks are modeled as triangle edge filters that approximate simple perceptual rules such as proximity, continuity, closure, etc. The filtering of triangle edges results in the simultaneous creation of regions and boundaries that are perceptually salient.