Although the linear scale-space representation generated by smoothing with the rotationally symmetric Gaussian kernel provides a theoretically well-founded framework for handling image structures at different scales, the scale-space smoothing has the negative property that it leads to shape distortions. For example, smoothing across “object boundaries” can affect both the shape and the localization of edges in edge detection. Similarly, surface orientation estimates computed by shape-from-texture algorithms are affected, since the anisotropy of a surface pattern may decrease when smoothed using a rotationally symmetric Gaussian.
KeywordsDiffusion Equation Gaussian Kernel Commutative Diagram Moment Matrix Shape Distortion
Unable to display preview. Download preview PDF.