Christoph Rasche

Gaze Control in Dynamic Broadband (1/f) Noise Sequences — by Christoph Rasche — last modified 2008-09-05 10:51

Research on gaze control has been typically characterized for stationary targets on homogenous backgrounds. In this study, gaze behavior is investigated with a dynamic broadband (1/f) noise display, which offers much more the kind of natural background distraction as when moving through natural scenes. To elucidate the possible bottom-up component of saccadic target selection, we analyzed the fixation locations during free viewing and find that looming dark spots are preferred and the ROC area value for a target/random patch distinction was 0.57. For a luminance-target search, this bottom-up component was overwritten but the target/random patch distinction was not larger (ROC area value also 0.57). Exploiting the principle of the classification image, it can be shown that saccadic decision time may not be fixed but rather depend on target properties. Saccadic orienting properties are in qualitative agreement with measurements obtained from a stationary, homogeneous background, but saccadic constant and variable error increased. A cued search reveals that despite the presence of the distracting background, attentional shifts take place to obtain identification judgments about targets appearing in the parafovea.

The Potential of Contour Grouping for Image Classification — by Christoph Rasche — last modified 2010-03-15 13:44

An image classification system is introduced, that is predominantly based on a description of contours and their relations. A contour is described by geometric parameters characterizing its global aspects (arc or alternating) and its local aspects (degree of curvature, edginess, symmetry). To express the relation between contours, we use a multi-dimensional vector, whose parameters describe distances between contour points and the contours' local aspects. This allows comparing for instance L features or parallel contours with a simple distance measure. The approach has been evaluated on two image collections (Caltech 101 and Corel) and shows a reasonable categorization performance, yet its future lies in exploiting the preprocessing to understand 'parts' of the image.