Detection and recognition of textual information from drug box images using deep learning and computer vision

The scope of this thesis work is to implement an OCR pipeline, capable of detecting and recognizing text instances when an image is given as input. The pipeline is divided into two steps: a detector, which scope is to detect the regions where a text is present, and a recognizer, which scope is to recognize and read the detected words and numbers. The work was initially developed during the internship experience in the start-up PatchAI, now an Alira Health company. The application of the algorithm in this context is the recognition of textual information on drug boxes. The idea is to deploy such pipeline into an app support, in such a way it can be used by patients, who can take a picture of the box and receive information about the medicine, in particular its posology. Also the use of a vocal assistant that reads orally the recognized text is explored, being a interesting application for ederly or visually impaired people.