Optical Character Recognition

The goal of this project is to provide a way of detection and recognition of texts within images following two main approaches; detection of texts, recognition of contents and simultaneous end-to-end text spotting. Our aim is to provide the capability for 6 unique use cases that Huawei’s onsite technicians face in their daily life such as applying OCR, extracting texts from an air conditioner and battery labels, site plates, RRU and antenna labels, electronic screens, ID cards, and meter indicators.

Our OCR solution will be incorporating multiple modules such as semantic segmentation, salient area detection, and text embedding on detectable areas into a single monolithic pipeline. Thus, it can be used to create an infinite amount of specific training data to our combined text spotting network, which will be the merged version of text detection and text recognition networks.

Apart from current targets, the following are also potential application areas;
1. Document Analysis
2. Scene Understanding
3. Autonomous Driving
4. Product Identification
5. Image Retrieval
6. Robot Navigation