Performance Comparison of Cosine, Walsh, Haar, Kekre and Hartley Transforms for Iris Recognition using Fractional Energies of the Transformed Iris Images
Sudeep D. Thepade, Pushpa R. Mandal · 2014
This paper presents a novel Iris feature extraction technique using fractional energies of transformed iris image. To generate image transforms various transforms like Cosine, Walsh, Haar, Kekre and Hartley transforms are used. The above transforms are applied on the iris images to obtain transformed iris images. From these transformed Iris images, feature vectors are extracted by taking the advantage of energy compaction of transforms in higher coefficients. Due to this the size of feature vector reduces greatly. Feature vectors are extracted in 5 different ways from the transformed iris images. First way considers all the higher energy coefficients of the transformed iris image while the rest considers 99%, 98%, 97%, and 96% of the higher energy coefficients for generating the feature vector. Considering fractional energies lowers the computations and gives better performance. Performance comparison among various proposed techniques of feature extraction is done using Genuine Acceptance Rate (GAR). Better Performance in terms of Speed and Accuracy is obtained by considering Fractional Energies. Among all the Transforms, Cosine and Walsh Transform gives good GAR value of 85% by considering 99% of Fractional Energy. Thus, using Fractional Energy gives better performance as compared to using 100% energies. The proposed technique is tested on Palacky University Dataset.