Many vision systems require text recognition of images captured at very high speed: car number identifiers, scanners, etc. Current technology allows the use of complex image preprocessing systems to improve their reading characteristics. This thesis proposes two new binarization algorithms suited to high precision character recognition applications.

The starting point is the study of the different existing voting systems and of their advantages and disadvantages. These different systems have been studied from the point of view of security, privacy and ease of use. The study of the different binarization systems and their usefulness in electronic voting applications has been a second step.

Once noticed that the existing systems do not completely solve the problem, two new algorithms have been developed and tested under the conditions imposed by the voting systems. Both new algorithms are based on the use of neural networks and modified histograms. One of the algorithms has been selected, based on the use of a semantic description of the histogram and a general regression neural network, since it is the best under the requirements imposed by the voting systems.

Finally, a physical implementation has been proposed using programmable logic devices. Using this physical implementation, the algorithm has been tested in terms of speed and complexity added to the voting system.