| Summary: Recently, appearance based recognition approach became an important research topic in pattern recognition area. Model based approaches are special to the problem and proper selection of extracted features are very difficult. Therefore, appearance based approaches become more effective and popular. Applications for recognition of biometrics, e.g. face, palm, iris; generic object recognition, medical recognition are being developed in order to partially or fully performed by appearance based approaches.
In this work, for better characterization of objects, usage of nonlinear approaches which provide higher order statistics is proposed. As a first stage, patterns are transformed to the spectral domain with feature decomposition, and redundant or noisy data are removed from this domain. Then, kernel concept which converts linear approaches to the nonlinear approaches without causing computational complexity is followed up with interest. Linear and nonlinear approaches for extracting features are compared. In the classification which is final stage of pattern recognition, in addition to the basic approaches, one of the current approaches SVM is utilized. Our proposed method is applied to the palmprint databases and experimental results show that this method is quite successful.
Keywords: Pattern Recognition, Kernel Methods, Principal Component Analysis, Support Vector Machines, Palmprint Recognition. |
