OCR: Análisis e implementación de algoritmos en nuevas tecnologías de paralelización

Abstract

In recent years it has emerged the need to analyze a lot of data. There is not only the need to analyze a lot of data, but also the calculations are getting more complex (weather simulations, cryptography or bioinformatics). These facts lead to a huge need for computing capacity. To face this problem, one of the most extended techniques is parallel computing. This technique requires a process of adaptation or modification of the traditional codes, which are executed serially. This transformation is called “parallelization”. The central axis of this project is “Parallelization”. It shows the efficiency of “parallelization” in the field of artificial vision, which requires a lot of data processing. Specifically, we will analyze the parallelism to speed Optical Character Recognition (OCR). We will see the effectiveness of "parallelization" on new technologies such as Smartphones and Single-Board computer (SBC). The algorithm used to implement OCR in this project is KNN (K-Nearest Neighbors). Furthermore, and analysis of parallelization of the SVM (Support Vector Machine) algorithm has been done. The KNN algorithm is going to be parallelize and its performance will be analyzed in the aforementioned platforms. Finally, the efficiency of parallelization is measured comparing execution times and power consumption, between parallel versions and their corresponding serially versions.