Abstract Effectively describing and recognizing leaf shapes under  arbitrary deformations, particularly from a large database,  remains an unsolved problem. In this research, we  attempted a new strategy of describing shape by walking  along a bunch of chords that pass through the shape to  measure the regions trespassed. A novel chord bunch walks  (CBW) descriptor is developed through the chord walking  that effectively integrates the shape image function over the  walked chord to reflect the contour features and the inner  properties of the shape. For each contour point, the chord  bunch groups multiple pairs of chord walks to build a  hierarchical framework for a coarse-to-fine description.  The proposed CBW descriptor is invariant to rotation,  scaling, translation, and mirror transforms. Instead of  using the expensive optimal correspondence based  matching, an improved Hausdorff distance encoded  correspondence information is proposed for efficient yet  effective shape matching. In experimental studies, the  proposed method obtained substantially higher accuracies  with low computational cost over the benchmarks, which  indicates the research potential along this direction