End-Stopped Wavelets for Detecting Low-Level Features

Certain kinds of cells, called end-stopped cells have been found to exist in the primary visual cortex of several mammals. Two kinds of end-stopped cells have been identiied. The single end-stopped cells respond strongly to corner-like stimuli in the real world. The second type of end-stopping is eeected by double end-stopped cells which respond strongly to a linear stimulus oriented in a speciic direction, and a speciic length. Double end-stopped cells respond to short linear segments, or strongly curved lines. End-stopping behavior is useful in selecting meaningful, low level image-features, which can be used in a variety of image-analysis problems. In this paper, we present two lters that simulate the behavior of biological end-stopped cells. Both are zero-mean lters, and are well localized in the spatial as well as frequency domains, that is, these lters are admissible wavelets. We refer to the two lters as ES1 and ES2. The ES1 lter responds to ends of linear structures which have a speciic orientation, and the ES2 lter responds to line-segments which have a speciic orientation, and which have a length within a speciic range. We show sample results to demonstrate the behavior of the proposed wavelets, and we also discuss the scale-space behavior of these wavelets brieey.