Smooth Splicing: A Robust SNN-Based Method for Clustering High-Dimensional Data

Algorithm 4

SNN similarity based smooth splicing clustering algorithm.

(1) Calculate SNN similarity graph.

(2) Identify the 𝐶²_𝑚 edges, remove the edges with zero intensity, and sort the rest edges according to their strengths in

  descending order.

(3) While edge set is not empty, then the smoothness is set as 1, starting from the first edge, study the following edges

   one by one whether they satisfy the 1 order splicing conditions, if

 (3.1) Yes,

    (3.1.1) Two edges splice;

    (3.1.2) Regard the newly joined vertexes as the splicing point, and make use of the property 2 to splice continuously,

          meanwhile smoothness gradually increases, until they do not meet the conditions for splicing;

    (3.1.3) Prune the edges containing the vertexes in layer 𝑘−1 from the edge set when the layer 𝑘 is completed;

    (3.1.4) When no edge in edge set meets the conditions for splicing, then prune the edges containing the outermost

         vertexes of the single-connected graph from the edge set, and go to 3.

 (3.2) No, start from the first edge in edge set, repeat 3.

(4) If edge set is empty, then each single-connected graph represents a category, where some small sample sets can sets can

  be regarded as noise category.