Line detection discrete scale-space

Based on the scale-space theory for discrete signals introduced by Lindeberg (1990) a line detector is presented for locating roof-shaped lines in 1-D and 2-D signals with subpixel accuracy. Additionally the slope of the profile and the line-orientation are estimated. This approach uses convolution kernels derived from a discretization of the diffusion equation in space and scale. The reasons for not using the scale-space operator with continuous scale parameter as proposed by Lindeberg are discussed. The approach features high accuracy and little computing effort for each level of scale