Signature Verification

Source: Deep Learning on Medium


Go to the profile of Anurag Kumar

In this article, I present the off-line signature verification system which is based on image processing, moment invariant method, and ANN. sequential neural networks are designed, and another for verification (i.e. for detecting forgery).

Image Processing

The scanned real world images containing human signatures are processed using several image processing algorithms before the calculation of the moment invariants. These processes are given below.

Converting color image to grayscale image:

In the present technology, almost all image capturing and scanning devices use color. Therefore, we also used a color scanning device to scan signature images. A color image consists of a coordinate matrix and three color matrices. The coordinate matrix contains x, y coordinate values of the image. The color matrices are labeled as red (R), green (G), and blue (B). Techniques presented in this study are based on grey scale images, and therefore, scanned or captured color images are initially converted to grey scale using the following equation :

Gray color = 0.299*Red + 0.5876*Green + 0.114*Blue

Noise reduction: Noise reduction (also called “smoothing” or “noise filtering”) is one of the most important processes in image processing. Images are often corrupted due to positive and negative impulses stemming from decoding errors or noisy channels. An image may also be degraded because of the undesirable effects due to illumination and other objects in the environment. A median filter is widely used for smoothing and restoring images corrupted by noise. It is a non-linear process useful especially in reducing impulsive or salt-and-pepper type noise. In a median filter, a window slides over the image, and for each positioning of the window, the median intensity of the pixels inside it determines the intensity of the pixel located in the middle of the window. Different from linear filters such as the mean filter, the median filter has attractive properties for suppressing impulse noise while preserving edges. A median filter is used in this study due to its edge preserving feature.

Background elimination: Many image processing applications require the differentiation of objects from the image background. Thresholding is the most trivial and easily applicable method for this purpose. It is widely used in image segmentation. We used the threshold technique for differentiating the signature pixels from the background pixels.

We used Adaptive Gaussian Thresholding

Skeletonization: The aim of the skeletonization is to extract a region-based shape feature representing the general form of an object. We used Zhang-Suen Skeletonization Algorithm

Moment Invariant Method

Moment invariants are properties of connected regions in binary images that are invariant to translation, rotation, and scaling. They can be easily calculated from region properties and they are very useful in performing shape classification and part recognition. One of the techniques for generating invariants in terms of the algebraic moment was originally proposed by Hu.

Feature vectors used for signature identification are generated using moment invariants. For this purpose, we produced twelve different sets of feature vectors for every signature where each set consisted of seven moments invariant. First, a feature vector for the normalized signature is produced using moment invariants, and these seven feature vector values are saved in a database. Then, the normalized signature is rotated using the two-dimensional rotation equation.

Twelve different sets of feature vectors are calculated to correspond to twelve different reference angles. The normalized signature image is rotated from 0 to 360 in 30 increments. In total, 84 features were extracted for each signature.

Hu moments for 1 image, similarly we will have 84 moment set for 1 Signature

Signature Verification

In this part of the study, our purpose is to authenticate a signature, i.e., to verify that the signature is not counterfeit and that it really belongs to the person who is claimed to be the owner of the signature. The ANN used for this purpose is also a multilayer feedforward network which consists of 84 input variables, 40 hidden neurons, and 2 output variables indicating whether the signature is fake or true. Backpropagation algorithm is used for training. The training data set is obtained from five original (authenticated) signatures provided by the real owner and five fake signatures. As was done for the preparation of the training data for the ANN used in recognition, twelve invariant vectors per signature are used in the training set.

Tested the verification network using 80 signatures: 40 imitations (counterfeit signatures) and 40 true signatures. The program detected (classified) 34 true signatures and 39 counterfeits correctly. In other words, almost all counterfeit signatures were detected correctly. Only six signatures were classified as counterfeits while they were not (i.e. “false negative”)

Conclusion

In this article, we presented an off-line signature verification system which is based on image processing, moment invariants, and ANNs. Moment invariants which are used as input features for the ANN are obtained from thinning signature images.

However, it exhibited poor performance when it was presented with signatures that it was not trained for earlier. We did not consider this as a “high risk” case, because the recognition step is always followed by the verification step and these kinds of false positives can be easily caught by the verification system. The verification system missed only one counterfeit signature.