YU Qinru1, LU Guifu2*, LI Hua2
(1 Wuhu Institute of Technology, Wuhu 241000, Anhui, China;2 School of Computer and Information, Anhui Polytechnic University, Wuhu 241009, Anhui, China)
Abstract:
In the GraphSC algorithm, the Laplacian graph is pre-defined and fixed, and will not participate in the subsequent learning process of the dictionary and sparse coding, the pre-defined Laplacian graph isnt the most suitable. Graph regularization sparse coding with adaptive neighbour algorithm (GraphSCAN) is proposed to solve the problem.The algorithm uses an adaptive method to construct a suitable local Laplacian graph, and then adds it to the SC objective function. GraphSCAN incorporates graph construction and sparse coding into a unified framework, so that graph construction and sparse coding operations are iteratively performed simultaneously. The experimental results of image clustering on CMU face data and COIL20 data support the effectiveness of the GraphSCAN algorithm.
KeyWords:
graph regularization; sparse coding; image clustering; adaptive clustering
