Binarisation system based on FPGA for OCR in electronic voting

Many vision systems require text recognition captured at very high speed: vehicle registration identification devices, scanners, etc. Current technology enables the use of complex image pre-processing systems to enhance reading characteristics. This PhD thesis puts forward two binarisation algorithms that are suitable for high-precision applications in optical character reading (OCR). The binarisation of a digital image involves the conversion of the digital image into a black and white one in such a way that the essential properties of the image are conserved. Most reading recognition algorithms are written based on binary images.

A study of the various types of current voting systems, their advantages and drawbacks, was taken as the starting point.

Disadvantages present in the various types of voting systems were taken to be those involving security and privacy when casting a ballot as well as the user-friendly nature of the system. A second stage was to study the various current systems of binarisation and assessing their suitability to electronic voting applications.

Once it was evaluated that the current systems do not completely solve the problem, two algorithms were drawn up and their operation under the conditions imposed by the voting systems was evaluated. The two put forward were based on the use of modified neuronal and histogram networks. One of the two algorithms was chosen, based on its use in the semantic description of the histogram and a general regression neural network, as it was the most suitable taking into account the characteristics of the voting systems.

Finally, a physical implementation for the chosen algorithm was proposed, based on programmable logic devices. By means of this physical implementation it was possible to evaluate that the algorithm was, in fact, suitable in terms of speed and added complexity for the voting system.