Deep Representation based feature extraction and recovering for Finger-vein verification
Finger-vein biometrics has been extensively investigated for personal verification. Despite recent advances in fingervein verification, current solutions completely depend on domain knowledge and still lack the robustness to extract finger-vein features from raw images. This paper proposes a deep learning model to extract and recover vein features using limited a priori knowledge. Firstly, based on a combination of known state of the art handcrafted finger-vein image segmentation techniques, we automatically identify two regions: a clear region with high separability between finger-vein patterns and background, and an ambiguous region with low separability between them. The first is associated with pixels on which all the segmentation techniques above assign the same segmentation label (either foreground or background), while the second corresponds to all the remaining pixels. This scheme is used to automatically discard the ambiguous region and to label the pixels of the clear region as foreground or background. A training dataset is constructed based on the patches centered on the labeled pixels. Secondly, a Convolutional Neural Network (CNN) is trained on the resulting dataset to predict the probability of each pixel of being foreground (i.e. vein pixel) given a patch centered on it. The CNN learns what a fingervein pattern is by learning the difference between vein patterns and background ones. The pixels in any region of a test image can then be classified effectively. Thirdly, we propose another new and original contribution by developing and investigating a Fully Convolutional Network (FCN) to recover missing fingervein patterns in the segmented image. The experimental results on two public finger-vein databases show a significant improvement in terms of finger-vein verification accuracy.
PROJECT OUTPUT VIDEO:
In existing system, the fingervein image, by contrast, is enhanced based on curvelets, and then a local interconnection neural network with a linear receptive field is employed to learn straight-line vein features based on labeled patches.
The network is trained to detect a horizontal line and the receptive field of the neural network is further rotated by an angle to extract other lines.
Unfortunately, no details on the experimental setup are given and the pixels seem to be manually labeled.
DISADVANTAGES OF EXISTING SYSTEM:
The approaches above extract finger-vein patterns based on attribute assumptions such as valleys and straight-lines. As a result, they suffer from following problems:
A) These assumptions are not always effective to detect the finger vein patterns. Compared to background pixels, the vein pixel values from clear regions do correspond to valleys or straigth-line attributes, so, in this case, the approaches above can distinguish vein from non-vein pixels. However, they cannot work well for ambiguous regions, because noise can comprise some valley structures in the vein region while creating false valleys in the background.
B) As the pixel values can create different distributions, it is impossible to model all attributes which are related to finger-vein features. Most of attributes used in current state of the art are those which are easy to observe and model, such as valleys or straight lines. Considering as many existing attributes as possible may alleviate the problem but is costly intensive, beside requiring heavy manual inspection.
C) It is not easy to propose a mathematical model to effectively describe the distributions shown, for instance, assuming the attributes are related to finger-vein patterns. The distributions are very complicated to model accurately by a mathematical model, as this is illustrated by the lack of modeling approaches on this problem.
We propose a deep learning model for finger-vein verification. Our approach aims,
first, at segmenting foreground (vein) pixels from background pixels by predicting the probability of a pixel to belong to a vein pattern given limited knowledge, and, second, at recovering missing vein patterns. Compared to current state of the art segmentation and recovering approaches, that are based on image processing techniques, our approach does not segment or recover an image based only on its pixels and their correlations, but it does so by relying also on rich statistics on nonlinear pixels correlations, through a hierarchical feature representation learned by a deep neural network from a large training set. This is a major advantage over traditional approaches as relying only on noisy input images for segmentation or vein recovery may lead to severe errors.
The main paper contributions are summarized as follows:
1) We propose an automatic scheme to label pixels in vein regions and background regions, given very limited human knowledge. We employ several existing baselines approaches to extract (segment) the vein network from an image and use
their combined output automatically to assign a label for each pixel. Such a scheme avoids the heavy manual labeling and may also reduce label errors, especially for ambiguous pixels.
2) A CNN-based scheme is employed to automatically learn features from raw pixels for finger-vein verification. First, a dataset is constructed based on patches centered on the labeled pixels, and we take the patches as input for CNN training. Secondly, in the test phase, the patch of each pixel is input into CNN the output of which is taken as the probability of the pixel to belong to a vein pattern. Then, the vein patterns are segmented using a probability threshold of 0.5. Compared to existing approaches, our CNN automatically learns robust attributes for finger-vein representation. Experimental results on large public datasets show that the proposed model is able to extract the vein patterns from raw images in a robust way, which leads to a significant improvement in finger-vein verification accuracy.
3) This paper investigates a new approach for recovering vein patterns in the extracted finger-vein image. As finger-vein patterns may be missing by corruption during the imaging stage and the inaccurate estimation of parameters during the preprocessing stage (i.e. alignment and feature extraction), we develop a robust finger-vein feature recovering scheme based on a Fully Convolutional Network (FCN). In this context, we perform a rigorous experimental analysis that shows that our scheme does succeed in recovering missing patterns which further improves the verification performance.
ADVANTAGES OF PROPOSED SYSTEM:
Experimental results show that the proposed approach extracts robust vein patterns and significantly improves the verification error rate w.r.t the state of the art.
Currently, the proposed approach achieves promising performance for finger-vein verification
System : Pentium Dual Core.
Hard Disk : 120 GB.
Monitor : 15’’ LED
Input Devices : Keyboard, Mouse
Ram : 1GB.
Operating system : Windows 7.
Coding Language : MATLAB
Tool : MATLAB R2013A
Huanfeng Qin and Mounim A. El Yacoubi, “Deep Representation based feature extraction and recovering for Finger-vein verification”, IEEE Transactions on Information Forensics and Security, 2017.