Research and application of variant granularity feedback recognition method based on maximal entropy and cloud-membership
Keqiong Chen, Jian‐Ping Wang, Weitao Li, Chenghui Zhu, Bo Hu Li · 2014
In order to improve the recognition performance and avoid the high dimensional feature space by using multi-feature fusion technology, indexes of recognition capability entropy, feature recognition intensity, recognition information entropy, and feature weight are firstly defined. Then, the multi-feature fusion algorithm based on maximal recognition capability entropy and feature selection algorithm based on feature recognition intensity and recognition information entropy are proposed to generate a compact feature set to enhance the generalization capability of the feature space. According to cloud model theory, multi-dimensional particle cloud model of extracted compact feature set is generated by multi-dimension reverse cloud generator. Finally, a variant granularity feedback recognition mechanism is designed to solve the coincident code problem by traditional cloud-membership maximal decision method and realize the variant granularity knowledge mining for the extracted compact feature set. The recognition of offline handwritten Chinese character is employed as the application object to verify the feasibility and effectiveness of our proposed method.