Scale-space Technique Based On K-curvature

The curvature scale-space (CSS) technique is considered to be a modern tool in thefiled of computer vision. Due to a number of excellent properties, CSS has been selectedinto the MPEG-7standardization. A CSS map can be obtained by an evolution of the inputcurve, where the trajectories of curvature zero-crossings are recorded during the change ofthe scale parameter. For a digital curve, it is first needed to convolve it with the Gaussian,and compute its curvature via a complex equation. Due to the complexity, CSS has a highcomputational cost. To overcome the difficulty, in the first part of the paper, we propose anew scale-space technique for shape description based on the k-curvature. First, thek-curvature of every point in the curve is computed, where the parameter of thek-curvature is treated as a scale parameter. As the scale parameter increases, the curvaturezero-crossings are extracted and recorded, whose trajectories construct a k-CSS map forshape description. Since there is no need of curve evolution, k-CSS has a lowercomputational cost and is easier to be coded. Experimental results show that k-CSSperforms well for shape description and analysis.In the second part of the paper, a shape match algorithm is proposed based on k-CSS.First, the arc-shaped contours in a k-CSS map are simplificated into rectangles accordingto their heights, and widths. This simplification can retain the strength of the curvaturefeature of the input curve, as well as the distance information betweens the feature points.As a result, the shape can be described more precisely when compared with the existingCSS algorithm. Finally, a shape matching algorithm is developed, and experimental resultsindicate it has a high consistency with the human beings’view system.