The International Conference on Document Analysis and Recognition (ICDAR) is an international academic conference which is held every two years in a different city. The ICDAR Robust Reading Competition has been held five times previously; in 2003, 2005, 2011, 2013 and 2015. The competition is organized around challenges that represent specific application domains for robust reading.
On October 17th, Dahua Technology took first place in Task [Word Recognition] of Incidental Scene Text Challenge and Born-Digital Image Challenge–with an accuracy of 82.76% and 97.43% respectively.
Incidental Scene Text
Incidental Scene Text, a new challenge to the 2015 edition of the competition, is the most difficult task. It refers to text that appears in the scene without the user having taken any specific prior action to cause its appearance or improve its positioning /quality in the frame. Incidental scene has strong relevance to a wide range of applications linked to wearable cameras or massive urban captures where the capture is difficult to control.
Born-Digital Image is one of the two challenges from the very first edition of the competition - ICDAR 2011. It refers to images saved by digital devices from the Internet and emails. Automatically extracting text from born-digital images is an interesting prospect as it would provide the enabling technology for a number of applications such as improved indexing and retrieval of Web content, enhanced content accessibility, content filtering (e.g. advertisements or spam emails) etc.
Dahua’s team developed
Robust Reading and ICDAR 2015
"Robust Reading" refers to the research area dealing with the interpretation of written communication in unconstrained settings. It has significant implication to video surveillance applications such as vehicle number plate recognition (ANPR), container serial number recognition, logistics label context recognition and the recognition of text captured in normal surveillance.
The Dahua AI OCR team from Dahua Advanced Research Institute participated in the ICDAR Robust Reading 2015 competition. Based upon deep learning technology and the advantages of SENet and ResNet network structure, Dahua’S team developed a unique strategy of multi-featured and multi-channel integration. Deployed together with multi-model integration technology, it greatly enhanced the accuracy of result.
The technologies utilised in this competition have been widely applied in Dahua smart transportation solutions. A breakthrough performance was achieved in situations dealing with tilted car number plates, with recognition rate reaching up to 99.99%.
In recent years, deep-learning has led to revolutionary breakthroughs in Intelligent Video Analytics. Accuracy of recognition is better than human perception in many situations. It becomes possible and economically viable to automate many tasks that were not possible before. AI has been widely applied in public security, transportation and banking to protect property and people. With a mission of “Enabling a safer society and smarter living”, Dahua will continue to focus on “Innovation, Quality, and Service” to serve partners and customers around the world.