Rank information fusion for challenging ocular image recognition

摘要

Under challenging imaging conditions which include lower resolution, occlusion, motion and de-focus blur, iris recognition performance degrades. In such conditions ocular region has been suggested as a new biometric modality which has the ability to overcome some of the above mentioned drawbacks. In this work, we investigate the performance of rank level fusion approach that fuses the outputs of three ocular region matching algorithms, namely, Probabilistic Deformation Model (PDM), modified Scale-Invariant Feature Transform (m-SIFT) and Gradient Orientation Histogram (GOH), employed for recognizing challenging ocular images in the Face and Ocular Challenge Series (FOCS) dataset. We investigate different rank fusion schemes including the highest rank, Borda count, plurality voting and Markov chain and demonstrate that rank-level fusion can lead to improved recognition performance.