Metodologia de processamento de imagens para leitura e reconhecimento digital seletivo de texto

Abstract

The objective of this work was to implement a text selection methodology based on the difference of affected color components in digital images after the use of text markers and the application in digital recognition of the selected text using an OCR library in MatLab Software. The activities were carried out in two digital images, one created using an editing software and the second obtained from a database, both marked with a digital text marker. MatLab image manipulation tools were used to identify the affected color component in the images and isolate this component in order to obtain a selection of the text in both images. For the ideal image, a satisfactory result was obtained, with the text selected and recognized by the OCR library. On the other hand, when applying the methodology in the image of a database, the noise present in the image made the selection difficult, making it impossible to read it by the OCR library. In an attempt to circumvent the flaw, the edge detection technique was used to better define the selected text. The results obtained with the ideal image were again satisfactory, but still without text recognition for the database image. The study provided a positive view on the selection of texts from markers, it is expected that in future works it will be possible to apply the technique in a more precise and automated way in order to obtain an accurate result for real and large-scale images.