Authors : Xinbo Gao , Jie Li and A. Halidan
Abstract: Cluster analysis is an important unsupervised machine learning technique and has been widely applied in the field of pattern recognition, data mining and computer vision. As a popular clustering method, fuzzy c-means algorithm partitions any a specified pattern sample set into clusters and reveals the degrees of sharing for each pattern sample in all the clusters. To find the meaningful clusters as defined by dense regions rather than to generate partition, the possibilistic c-means (PCM) was proposed as a mode- seeking algorithm to cast the clustering problem into the framework of possibility theory. The PCM algorithm assigns and interprets the membership degree as typicality or compatibility. However, each clustering prototype obtained by the PCM algorithm is the possibilistic mean of all the patterns in the corresponding cluster, which sometimes does not correspond to a real existing pattern in the data set. To extract the real typical pattern as clustering prototype for each cluster, a novel possibilistic c-patterns (PCP) algorithm is presented in this paper. Like the PCM algorithm, the PCP algorithm is also a mode-seeking clustering algorithm. However, it can be used to search the representative samples as prototypes for all the obtained clusters. Based on the favorable characteristics of the PCP algorithm, it is applied to video abstraction for content-based video retrieval. The experimental results with synthetic data and real video data illustrate the effectiveness of the typical-pattern seeking from noise contaminated data set as well as the key frame extraction of video sequence.
Xinbo Gao , Jie Li and A. Halidan, , 2005. A Possibilistic C-Patterns Algorithm and its Application in Video Abstraction . Asian Journal of Information Technology, 4: 61-67.