Abstract
Purpose: Object recognition requires that one reconstruct aspects of a 3-Dimensional scene from its 2-Dimensional image. More specifically, the observer must explain image contrast patterns in terms of their causes in the environment. What features in the image facilitate this estimation? We hypothesized that edge boundaries, texture, repetition of elements, and edge profile would be predictive of human cause estimations.
Methods: In study one, subjects were asked to view regions of interest (ROIs) containing an edge and choose one of four possible edge causes (i.e. albedo, depth, shadow, specularity). In the second part of the study, a group of trained “experts” were asked to look at the same ROIs and determine whether or not each contained particular features (i.e. closure, texture, or repetition). Finally, a clustering algorithm was used to classify the ROI's edge profile.
Results: Analysis for expert feature classification was done using a 3×3×2 MANOVA with image features as factors and cause estimation as the dependent variables. The analysis determined that edge image features have an effect on edge cause estimation. There were main effects of all three factors. Closure had an effect on albedoness, depthness, and specularness (pConclusion: Closure, texture, repetition of elements, and edge profile are predictive of human edge causation estimation.
This publication was made possible by Grant Number P20-RR-020-151 from the National Center for Research Resources (NCRR), a component of the National Institutes of Health (NIH).