Micro-dimensional Feature Extraction for multispectral Iris recognition

Abstract

Feature extraction of multispectral iris images relies heavily on location and dimensions to effectively extract the most relevant features. For recognition, iris images are normally captured in the 700nm to 900nm range. This is due to the fact that ranges in the normal 350-700nm range cannot so easily penetrate darker colored irises to get texture information, whereas larger ranges can. There have been a number of techniques used for iris based feature extraction. These techniques have been focused on reducing the amount of features needed while preventing a reduction in accuracy. Techniques have been envisioned that combine artificial intelligence techniques with these feature extraction techniques. One such technique is Genetic and Evolutionary Feature Extraction (GEFE); GEFE evolved the optimal location and dimensions for feature extraction techniques. A variation of GEFE, known as micro GEFE (mGEFE), restrained the dimensions of areas to a pixel size of 4 by 4. This was an exploratory data analysis approach that pin pointed the most salient features on images. This paper proposes a variation that is built upon GEFE and mGEFE referred to as micro dimensional GEFE (mdGEFE). This technique will increase the accuracy from GEFE while using fewer features. Results show that mdGEFE has a comparable performance in accuracy with an approximate 50% reduction in surface area used.