Offline text-independent writer identification using codebook and efficient code extraction methods
Image and Vision Computing
| Hi-index | 0.00 |
Automatic writer identification systems have several applications for police corps in order to perform criminal and terrorist identification. The objective is to identify individuals by their off-line manuscripts, using different features. Character level features are currently the best choice for performance, but this kind of biometric data needs human support to make correct character segmentation. This work presents a new system based on using Connected Component level features, which are close to character level and can be easily obtained automatically. Our experiments use Support Vector Machines using Connected Components gradient vectors to identify individuals with good results. A real-life database was used with real forensic cases in different writing conditions.