Parametric Feature Detection

  • Authors:
  • Simon Baker;Shree K. Nayar;Hiroshi Murase

  • Affiliations:
  • Department of Computer Science, Columbia University, New York, NY 10027, U.S.A.;Department of Computer Science, Columbia University, New York, NY 10027, U.S.A.;NTT Basic Research Laboratory, Morinosato Wakamiya, Atsugi-shi, Kanagawa 243-01, Japan. E-mail: murase@siva.ntt.jp

  • Venue:
  • International Journal of Computer Vision
  • Year:
  • 1998

Quantified Score

Hi-index 0.00

Visualization

Abstract

Most visual features are parametric in nature, including,edges, lines, corners, and junctions. We propose an algorithm toautomatically construct detectors for arbitrary parametric features.To maximize robustness we use realistic multi-parameter featuremodels and incorporate optical and sensing effects. Each feature isrepresented as a densely sampled parametric manifold in a lowdimensional subspace of a Hilbert space. During detection, thevector of intensity values in a window about each pixel in the imageis projected into the subspace. If the projection lies sufficientlyclose to the feature manifold, the feature is detected and thelocation of the closest manifold point yields the feature parameters.The concepts of parameter reduction by normalization, dimensionreduction, pattern rejection, and heuristic search are all employedto achieve the required efficiency. Detectors have beenconstructed for five features, namely, step edge (five parameters),roof edge (five parameters), line (six parameters), corner (fiveparameters), and circular disc (six parameters). The results ofdetailed experiments are presented which demonstrate the robustnessof feature detection and the accuracy of parameterestimation.