i'm working on tesseract 4.00.00 alpha,i need to know how does it work. Is it based on dictionary?how does tesseract extract features and how can i measure accuracy?