Distance Images and Intermediate-Level Vision
Early vision is dominated by image patches or features derived from them; high-level vision is dominated by shape representation and recognition. However there is almost no work between these two levels, which creates a problem when trying to recognize complex categories such as “airports” for which natural feature clusters are ineffective. We argue that an intermediate-level representation is necessary and that it should incorporate certain high-level notions of distance and geometric arrangement into a form derivable from images. We propose an algorithm based on a reaction-diffusion equation that meets these criteria; we prove that it reveals (global) aspects of the distance map locally; and illustrate its performance on airport and other imagery, including visual illusions.
KeywordsScale Space Image Patch Edge Density Edge Pixel Distance Image
Unable to display preview. Download preview PDF.
- 2.Fidler, S., Leonardis, A.: Towards scalable representations of object categories: Learning a hierarchy of parts. In: CVPR. IEEE Computer Society (2007)Google Scholar
- 10.Florack, L., ter Haar Romeny, B., Viergever, M., Koenderink, J.: The gaussian scale-space paradigm and the multiscale local jet. 18, 61–75 (1996)Google Scholar
- 12.Courant, R., Hilbert, D.: Methods of mathematical physics, vol. 2. Interscience (1962)Google Scholar