Studies on biological vision systems have revealed their amazing capabilities in real-time scene categorisation and object detection. Stemmed from some recent computer vision research, we experimented with feature selection on some feature schemes commonly used for natural vs man-made scene classification. A few simple attributes derived from fractal statistics were also introduced and found to be quite effective. By comparing our results obtained from ordinary digital images and those of high dynamic range image data, we questioned whether we are using attributes extracted from digitisation artifacts as 'useful' features for image analysis. Biological systems without digitisation may do it totally differently?
Last modified: Wednesday, 15-Jul-2009 12:30:17 NZST
This page is maintained by the seminar list administrator.