Professor Jin Lianwen' s Team Received Two Prize in the ICDAR17 MLT Competition

In the jus ended three MLT (multi language scene text detection and classification of languages) competitions in 2017 ICDAR International Conference in Japan, the team led by Professor Jin Lianwen in our school won the first place in end- to-end scene text detection and classification tasks (achieving obvious advantage in the later one), and won the second place with only 0.4 percent of the weak gap in the task of language classification.

ICDAR MLT Challenge is scene text detection and language classification challenge with the most languages with large scale of data, including Chinese, Japanese, Korean, English, French, Arabic, Italian, German and Indian. There are three tasks in the ICDAR MLT competition. Task one is to locate all kinds of text (with 9.5 frames per picture on average). The two task is to classify languages for words segmented in pictures. Task three is the end- to-end detection and classification for text in scene. ICDAR MIT images are collected from various scenes, in which the length, form, size and color of words vary, and contains a lot of real scene noise, including light, occlusion, tilt, text stacking, text mosaic, perspective changes and so on, which makes it quite challenging. 16 team from colleges and enterprises in different countries and regions took part in his year's competition.