Blob Features

The images below demonstrate blob feature generation. The leftmost image is the input image. The second image is a region label image obtained from a pyramid algorithm with shape constraints. The third image is a blob approximation of the regions after merging regions, again with shape constraints. The rightmost image shows a reprojection of blob colours onto the regions they represent. Green pixels are unassigned.

• The algorithm for blob feature extraction is described in the thesis:
  Low and Medium Level Vision using Channel Representations
• The robustness of the features and their application in view matching are discussed in the paper:
  P-E Forssén and A. Moe. View Matching with Blob Features.
  2nd Canadian Conference on Robot Vision, Victoria, BC, May 2005.
  
• An earlier, grey-scale only version of the blob extraction method is described in the paper:
  P-E Forssén and G.H. Granlund. Robust Multi-Scale Extraction of Blob Features.
  SCIA, Scandinavian Conference on Image Analysis, Gothenburg. June-July 2003.
  See also the on-line workshop proceedings: LNCS vol 2749.
• A C-implementation of the algorithm can be downloaded from my software page.