| Summary: | The identity of a person can be identified by analyzing biometric identification. Iris is one of the biometric that widely used in the field of security due to its uniqueness.
There are a lot of feature extraction methods and classification methods for iris classification. Classic local binary pattern (LBP) is one of the most useful feature
extraction methods. Moreover, K-Nearest Neighbour (K-NN) classifier is one of the widely use classifier due to its simplicity. Due to the current methods in feature extraction
are still improving, this project proposed a new feature extraction method to increase the performance of iris classification. In this project, a classification system is proposed with the one-dimensional local binary pattern algorithm (1D-LBP) with the K-Nearest Neighbour (K-NN) classifier and the system is developed by using a Raspberry Pi 3.There are eight different subjects used to classify in this classification system and each subject consists of seven samples of normalized iris image as input to the system. There
are two stages in the proposed classification system. Firstly, the 1D-LBP algorithm is used to extract the features of the normalized iris images and save the data in a text file
according to the subject and the combinations to evaluate for the next stage. Secondly, the K-NN classifier is used to classify the 1D-LBP based features from the first stage.
There are two methods to evaluate the features, which are one versus one and one versus many. Twenty-eight pairs of subjects are saved in different text files and classified under
one versus one method. There are twenty pairs of the subjects are achieved 100% of classification accuracy. There are seven combinations of the subjects are classified by using the one versus many method. The best performance of the one versus many is when the data cluster involves three classes. The accuracy is 100%. The classification accuracy is decreased when the number of subjects in the test data is increased. The performance of the one versus many classification is affected by the 1D-LBP based information and the value of K in K-NN classifier. In conclusion, the 1D-LBP algorithm is performance
well with K-NN classifier.
|
|---|