Research and Development of License Plate Recognition System Technology - Tigerwong

Business card scanning software of license plate recognition system in many smart phones has this function. On November 20, 2018, optical character recognition (hereinafter referred to as OCR) refers to the process of analyzing images to obtain text information. The test set of international document analysis and recognition Conference (ICDAR), an important indicator in the field of OCR, was refreshed again. Chinese universities and enterprises ranked among the top five in ICDAR 2015, followed by pixel anchor Nanjing University and Nanjing University of technology (psenet masktext, fot) Alibaba (inceptext) focuses on text recognition and application. In order to improve the text detection and recognition level of natural scenes, the International Conference on document analysis and recognition is one of the professional conferences organized by the international pattern recognition Society (IAPR). The International Conference on document analysis and recognition (ICDAR established a robust text reading competition in 2003 (robustradecompetition has so far involved more than 3500 teams from 89 countries.

Google microsoftamazon Facebook, Peking University, University of science and technology of China, Tencent, Sogou, etc. have participated in it. Results can be submitted at any time, and the test data set of robust text reading competition has been published online. The test data set of testing algorithm in natural scene text detection industry. ICDAR 2015 and icdar2017mlt are two test subsets. Cdar2015 and icdar2017mlt are authoritative data sets in the field of natural scene text detection.Surging news interviewed cloud Cong technology, which currently ranks first in the icdar2015 list. Li Yuan, an algorithm engineer of the Institute of science and technology, introduced it. Basically, all articles will be detected on it "is also the basis of the ranking. Cdar2015 list (November 20, 2018) the value of F represents the harmonic average of accuracy and detection rate. The larger the value of F, the better the detection result.

Precision represents the proportion of correct text in the detection result; detection rate (recal reflects the ratio between the detected text and the total number of picture text, which explains the list for the reporter. It can reflect whether the text has been missed. In order to detect the effect of the algorithm, the license plate recognition system needs to comprehensively consider the accuracy and detection rate, and the F value just represents the harmonic average of the accuracy and detection rate, which is also the basis for ranking. The former does not have the function of recognizing text, so it needs to be noted It means that natural scene text detection mentioned here is not equivalent to text recognition. It means to detect text in pictures.

On November 20, 2018, Li Yuan said. Yuncong technology published the latest paper effect on the preprint website arXiv, which proposed a pixel anchor framework for natural scene text detection. Before the paper was published, it was released by the International Conference on document analysis and recognition The pixel anchor algorithm was tested on icdar2015 and icdar2017mlt data sets, and the best results of icdar2015 were refreshed. It is reported that the icdar2015 pure English text detection data set. Icdar2017mlt contains nine languages such as Latin, English, Chinese, Korean, Japanese and Arabic. Yuncong's pixel anchor algorithm ranks fourth in the icdar2017mlt comprehensive list and the top three countries No public papers have been submitted by foreign institutions.

Natural scene text detection can be applied in a broader field, but compared with traditional text detection, OCR text detection and recognition in various commodities, sets or natural scene pictures in natural scenes face complex background interference, text blur and degradation, unpredictable lighting, font diversity, vertical text and inclined text The license plate recognition system is compared with the conservative OCR for high-quality document images, such as photo analysis, license plate recognition, image advertising filtering, scene understanding, commodity recognition, street view positioning, bill recognition, etc.