Loci Features and Neural Network for Handwritten Text Recognition

The handwritten text recognition system has received considerable attention due to its indisputable importance. To achieve this objective, A broad spectrum of approaches for both feature extraction and classification are utilized. In this work, the performance of features extraction techniques is analyzed, with a particular focus on the Loci features method. This method has demonstrated its robustness and effectiveness in character recognition, outperforming other techniques such as zoning and profile projection. Subsequently, a study on the structure of the multilayer perceptron neural network is conducted, achieving a very high accuracy using Loci features. In contrast, the accuracy does not exceed 90% for zoning and profile projection.