Feature extraction using sequential cumulative bin and overlap mean intensity for iris classification

Authors: AHMAD NAZRI ALI, SHAHREL AZMIN SUANDI, MOHD ZAID ABDULLAH

Abstract: This paper examines an approach generalizing a variant of the local binary pattern (LBP) method for iris feature extraction. The proposed method employs two different LBP variants called the sequential cumulative bin and overlap mean intensity for projecting the one-dimensional local iris textures into a binary bit pattern. The assigned bit, either 1 or 0 as a bit code, replaces the original intensity value using a specific condition for the respective reference element. The ratio value from the total transition of 1 to 0 along the row axis represents the feature of each iris image. The extraction only utilizes a small area of interest on the iris image that covers parts of the iris textures with minimum eyelid and eyelashes. The assessment employs the support vector machines classifier and the result demonstrates a good classification rate with average accuracy of 94.0% for the individual mode. However, the classification rate has improved to reach 96.5% accuracy if the assessment uses a concatenated mode set of features. Besides that, increasing the amount of samples in the training data by using the synthetic together with the original samples has also been able to improve the classification rate.

Keywords: 1D-Local binary pattern, histogram equalization, support vector machines, iris classification

Full Text: PDF