Разпознаване на текстова информация в графични обекти чрез дълбоки невронни мрежи
Recognition of textual information in graphic objects through deep neural networks

Summary/Abstract: Text-to-image recognition is a research area that is trying to develop computer systems with the ability to automatically read text from images or video with capabilities even in online mode. Conditionally, there are two tasks that are solved when recognizing text - its discovery and the second task is the conversion into text format. In the experiments, a dataset was created with about 95,000,000 images generated with text with different font sizes, using all words taken from Bulgarian and English dictionaries. A deep neural network built with a structure based on "Handwritten Text Recognition with TensorFlow" with an extended model by the authors, created with the help of TensorFlow and Python, which recognizes the symbols and their order found in graphic objects. The Python implementation of EAST: An Efficient and Accurate Scene Text Detector is used to detect places in graphic objects where there is text.

• Details
• Contents