Off-line Writer Verification Research Articles

Off-line writer verification based on simple graphemes

A method to writer verification based on handwritten stroke analysis is presented. The proposed descriptors correspond to an estimation of the pressure applied when writing using the grayscale image of the stroke. These descriptors are obtained from individual and simple graphemes, in contrast with the complexity of the handwritten stroke used in the signature processing systems. In addition, a study is presented which suggests that the combination of descriptors of simple characters improves the recognition capacity of the method. The descriptors considered correspond to different accuracy degrees of pressure distribution representation. Specifically, from the simplest representation to a more complex one, the descriptors proposed are as follows: the width of the stroke, the gray level of the grapheme skeleton, the average of the gray levels on the perpendicular line to the skeleton, and the approximation transformation coefficients of the area of the grapheme. The advantage of these descriptors is that they are invariant to scale and rotation. The descriptors performance was assessed using the original images and also reduced versions based on traditional methods such as Principal Component Analysis and Discrete Cosine Transform. For the evaluation, a one-vs-all scheme was considered which is consistent with the problem of identity verification. It was implemented with Support Vector Machine classifiers trained with K-Fold Cross Validation. The efficient search of SVM hyperparameters was performed with the heuristic optimization algorithm Simulated Annealing. The evaluation of individual simple characters gives a high average of hits and the combination of characters even improves the performance, getting closer to 100% of hits in identity verification. Qualitative and quantitative comparison with other methods and descriptors has been also carried out with satisfactory results.

Towards robust writer verification by correcting unnatural slant

Slant is a salient feature of Western handwriting and it is considered to be an important writer-specific feature. In disguised handwriting however, slant is often modified. It was tested whether slant is indeed an important factor and it was tested whether the distorting effect of deliberate slant change can be countered by a simple shear transform. This was done in two off-line writer verification experiments in image processing conditions of slant elimination and slant correction. The experiments were performed using three features based on statistical pattern recognition, including the state-of-the-art features Fraglets and Hinge. A new public dataset was created and used, containing natural and slanted handwriting by 47 writers. A striking result is that the average natural slant value is much less important for biometric systems than is usually assumed: eliminating slant yields just a 1–5% performance loss. A second result is that the effects of deliberate slant change cannot be fully countered by a simple shear transform: it raises performance on the distorted handwriting from 53–68% to 64–90%, but this is still lower than normal operation on natural handwriting: 97–100%.

Off-line writer verification utilizing multiple neural networks

A writer verification system based on multiple neural network classifiers is described, aimed to be easily extendable. Each writer reg- isters with the system by writing a set of discrete Chinese characters. One neural network is constructed for each enrolled character class on a per-person or a per-subgroup basis. The decisions from individual net- work classifiers are combined by voting. The method has been verified to work reliably on our testing database: a verification rate above 96% is achieved on short-term data. © 1997 Society of Photo-Optical Instrumentation En- gineers. (S0091-3286(97)01211-7)

Off-line writer verification using ordinary characters as the object

A system of writer recognition using ordinary characters as the verification object is proposed. In the system, some optional constants, the trimming number of a modified Mahalanobis' distance, the threshold coefficient and the number of characters in the text, are determined. The leave-one-out method is proposed for determining the optional constants. The feasibility of the proposed system was examined with an experiment using a database composed of Japanese writing. The effectiveness of two types of features (the direction feature and the are feature) was also compared. Error rates as low as 0.5% were achieved in the proposed system using the direction feature.