Chord transformation and the description of chord space feature are studied in detail, and the concrete approach for extracting geometric feature is achieved experimentally. Meanwhile，the robustness of this feature is proved through experiment. The feature has invariant character for translation, rotation and scale, and some degree of tolerance for affine transformation. On the other side, after filtering original image with the feature of a specifically shape of model in the chord space by inverse chord transformation, the result holds the edges of objects having the same or similar shape with the model while restraining other edges. Experimental results show that it is effective to use the feature for the similarity match between (geometric) shapes. Moreover, the approach is easy to be understood which can be implemented in various fields such as object recognition, target tracking, etc.