“School of Cognitive”

Back to Papers Home
Back to Papers of School of Cognitive

Paper   IPM / Cognitive / 9583
School of Cognitive Sciences
  Title:   Evolutionary Design of Gabor Filters with Application to Writer Identification
1.  Ali Borji
2.  Caro Lucas
  Status:   In Proceedings
  Proceeding: Proceeding of 12th International CSI Computer Conference CSICC 07
  Year:  2007
  Supported by:  IPM
Previous studies have shown the advantages of Gabor filters for the purpose of writer identification. Current study attempts to show that designing a set of such filters is domain specific however tuning of their parameters is not an easy task. Here we use genetic algorithms as a systematic approach to search in the space of filter parameters to choose a set of best filters for the problem of offline writer identification based on handwriting. Specifically, each chromosome encodes the parameters of filters and then during an evolutionary process those filters which show the best classification performance are selected. Identification was performed using two different classifiers: weighted Euclidean distance (WED) and nearest neighbor classifier (KNN). Our results show the superiority of proposed method compared with traditional Gabor filters and Gray Scale Co-occurrence matrices (GSCM).

Download TeX format
back to top
scroll left or right