Comparison of denoise algorithms to optimize hand vein pattern recognition

• Main Author: Surabhi Batra
• Other Authors: Li Fang
• Format: Final Year Project
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Computer science and engineering; DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: http://hdl.handle.net/10356/59916
• Institution: Nanyang Technological University

Biometric identification using palm veins has received substantial attention in recent years by researchers. This is because vein structures do not alter over the life span of an individual and are therefore, quite reliable. Typically, the process requires vein images to be captured in a digital format. However, during capture or transmission, the images often get degraded by noise. Image denoising aims to restore the image to its original state as far as possible without loss of information or addition of unnecessary details. The process of denoising remains challenging despite the availability of a plethora of algorithms in the spatial, transform and learning-based domains. Each algorithm has its own benefits and drawbacks. This report gives an overview of the different categories of denoising techniques and provides a comparison of noteworthy noise reduction algorithms on a database of 10 Near Infra Red vein images. It then proposes the technique most suited to vein images after conducting comprehensive experiments. The vein images obtained for experimental purposes are noisy. Therefore, the absence of reference or ideal images makes it challenging to define a qualitative benchmark. For the same reason, the report goes on to examine metrics to assess image quality in the reference and no-reference image domain. It then recommends the most appropriate one to determine the quality of the vein images. The extensive study in this report could serve as a useful reference in tracking and stimulating further research in vein image quality enhancement for biometric applications.