A study on filtering methods for dorsal hand vein images
Loading...
Date
2017-06
Authors
Choong, Jing Pei
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Dorsal hand vein imaging is gaining attraction from different fields due to its uniqueness. It is impossible to acquire noiseless hand image. Hence, researchers put a lot of efforts to propose noise filtering algorithms to filter out noise in hand images for further processing. However, there is still no specific study which has justified a filtering method that can perform effectively on dorsal hand vein image. Therefore, this research works is set-up to study on various filtering methods for dorsal hand vein images. Firstly, this study models the noises to generate noisy images for experiments. Two types of noises: Salt-Pepper noise and Gaussian noise are modelled as the real-world noises. Then noise filtering algorithms are applied. Typical and adaptive noise filtering methods have been studied and applied to filter noise in two samples of hand images. The choices of filtering methods are based on its past achievements. Different samples of noisy images with different composition of noise signal are used to test the performance of filtering methods out based on the quantitative measure and visual quality. Then, evaluation on the performance of filtering methods on dorsal hand vein images is made. After that, comparison and contrast are made based on the performance of the filtering methods on different cases. The evaluation results showed that Three Value Weighted Filtering is robust on filtering Salt-Pepper noise while Non-Local Mean Filtering and Adaptive Measure of Medium Truth Degree Filtering have the lowest performance on low and high Salt-Pepper densities respectively. At 0.1 Salt-Pepper density, TVWF achieved MSSIM at 0.9919 while NLMF and AMMTDF achieved MSSIM at 0.8691 and 0.9792 respectively. At 0.9 Salt-Pepper density, TVWF achieved MSSIM at 0.9275 while NLMF and AMMTDF achieved MSSIM at 0.8665 and 0.0373 respectively. On the other hand, Block Matching with 3D Filtering is slightly ahead in filtering Gaussian noise, it achieved MSSIM at 0.9696 and 0.9605 at 0.001 and 0.005 Gaussian variances respectively. TVWF has the lowest performance on filtering Gaussian noise. It achieved MSSIM at 0.5326 and 0.1891 for 0.001 and 0.005 Gaussian variances respectively. Besides, MSSIM shows its robustness in evaluation among other evaluation parameters.