days ago, Ali mother image team OCR (text recognition diagram) ICDAR Robust Reading technology to refresh the data set of the world’s best competition results, and significantly exceeded second. Robust Reading competition is currently the world’s most influential OCR technology competition. This is following the Alibaba image search leader of China Xiansheng ACM (ACM) "outstanding scientists" title, Ali’s image recognition technology once again boarded the top international stage.
(Figure ICDAR Robust Challenge 1 end-to-end official score)
with this leading OCR technology, Ali mother image team is able to identify the figure of 95% in the high accuracy of illegal text messages, effectively filtering malicious business promotion, safeguard the interests of consumers. 2015, Ali mother cumulative shield 46 million malicious promotion.
OCR is a classical problem in the field of computer vision, which has been concerned by the academic and industrial circles for a long time. In the industrial sector, Google, Microsoft, Amazon and other large Internet Co, and Tang technology, as the technology Kuang (Face++, Image++) and other computer vision technology startups in OCR technology for many years. As technology continues to mature, OCR also began to gradually use the Internet and other industries.
, according to Ali mother orientation algorithm technology, who is responsible for the cover Kun introduced in the Alibaba platform marketing ideas, goods are displayed in the form of pictures, image technology is an important means to understand the information. At the same time, there are unscrupulous businessmen in the picture with some illegal information to achieve the purpose of promoting the malicious, illegal picture text is a kind of large proportion, but the traditional monitoring method with artificial eye to review, time-consuming, especially with the growing number of pictures, this has become almost impossible task. In response to this strong demand, from the beginning of 2014, Ali mother image team began to focus on OCR technology, through the machine vision way to identify the text from the picture, thereby identifying the illegal copy of the information.
definition of malicious promotion, gaikun said, on the one hand, the relevant national laws and regulations, for contraband, pornography, superstition, fake, false, exaggerated breast weight loss type of violation punishment and management at different levels, on the other hand, there are higher requirements in laws and regulations to damage consumer platform interests or unfair competition promotion type actively trying to control. Among them, the most closely related with the consumer fake and shoddy single violation type, was Ali mother as the focus of the review object.
compared to other limited scenarios OCR applications, Ali mother to deal with creative pictures and merchandise Map >