Diversity-performance relationship in a handwriting recognition system based on bit-plane decomposition

The success gained by applying bit-plane decomposition methods to handwriting recognition have been demonstrated in our previous work [S. Hoque et al., 2003, 2002]. In this paper we address the relationship between the diversity and the improvements obtained by applying multiple combinations of various layers. These layers are obtained by applying a method based on an n-tuple based classification system, namely, the random decomposition technique proposed in S. Hoque et al. (2003). We investigated 5 combination methods and 9 diversity measures using data extracted from the NIST database. Results presented in this paper support the use of the bit-plane decomposition approach as a diversification method. Strong correlation was found between both the accuracy and the improvements and the diversity measures in the majority of the combination methods investigated.

[1]  P. Sneath,et al.  Numerical Taxonomy , 1962, Nature.

[2]  Ludmila I. Kuncheva,et al.  Relationships between combination methods and measures of diversity in combining classifiers , 2002, Inf. Fusion.

[3]  Herbert Freeman,et al.  Computer Processing of Line-Drawing Images , 1974, CSUR.

[4]  Kagan Tumer,et al.  Error Correlation and Error Reduction in Ensemble Classifiers , 1996, Connect. Sci..

[5]  Michael C. Fairhurst,et al.  Genetic Algorithms for Multi-classifier System Configuration: A Case Study in Character Recognition , 2001, Multiple Classifier Systems.

[6]  Fuad Rahman,et al.  An Evaluation Of Multi-Expert Configurations For The Recognition Of Handwritten Numerals , 1998, Pattern Recognit..

[7]  Michael C. Fairhurst,et al.  Bit plane decomposition and the scanning n-tuple classifier , 2002, Proceedings Eighth International Workshop on Frontiers in Handwriting Recognition.

[8]  Thomas G. Dietterich.  An Experimental Comparison of Three Methods for Constructing Ensembles of Decision Trees: Bagging, Boosting, and Randomization , 2000, Machine Learning.

[9]  Padraig Cunningham,et al.  Diversity versus Quality in Classification Ensembles Based on Feature Selection , 2000, ECML.

[10]  Fabio Roli,et al.  Design of effective neural network ensembles for image classification purposes , 2001, Image Vis. Comput..

[11]  Ludmila I. Kuncheva,et al.  Measures of Diversity in Classifier Ensembles and Their Relationship with the Ensemble Accuracy , 2003, Machine Learning.

[12]  Derek Partridge,et al.  Software Diversity: Practical Statistics for Its Measurement and Exploitation | Draft Currently under Revision , 1996 .

[13]  Simon M. Lucas,et al.  Recognition of chain-coded handwritten character images with scanning n-tuple method , 1995 .

[14]  Ron Kohavi,et al.  Bias Plus Variance Decomposition for Zero-One Loss Functions , 1996, ICML.

[15]  Jiri Matas,et al.  On Combining Classifiers , 1998, IEEE Trans. Pattern Anal. Mach. Intell..

[16]  G. Yule,et al.  On the association of attributes in statistics, with examples from the material of the childhood society, &c , 1900, Proceedings of the Royal Society of London.

[17]  Michael C. Fairhurst,et al.  A new chain-code quantization approach enabling high performance handwriting recognition based on multi-classi .er schemes , 2003, Seventh International Conference on Document Analysis and Recognition, 2003. Proceedings..