On August 8, MyScript won the international Competition on Online Handwritten Text Recognition. The contest was organized for the 16th International Conference on Frontiers in Handwriting Recognition (ICFHR 2018), in Niagara Falls, USA.
The ICFHR is one of the most renowned organizations of its kind, along with the International Conference on Document Analysis and Recognition (ICDAR). There are the only two bodies that deliver hallmarks of excellence in intelligent document recognition technologies.
Since 2006, MyScript has participated in 10 of their handwriting recognition competitions and always took first place.
The ICFHR is the ultimate point of convergence for handwriting recognition researchers and developers. Experts from academia and industry gather every other year for four days of intensive conferences, panel discussions, poster sessions and competitions. They debate on their research projects and the latest technologies they put together. The ICFHR covers fascinating technical topics such as:
This year’s challenge was on Vietnamese and competitors were to tackle 3 tasks: the recognition at word level, line level and paragraph level. MyScript completed the 3 assignments and obtained the best results in all of them. The most impressive win was at the line level, with an outstanding error rate of only 3% against 21% for the second-best result. At the other two levels, we also outperformed our competitors. For instance, at the paragraph level, we had an error rate of 6% against 24% for the contestant that took second place.
MyScript obtained the best results
Vietnamese is one of the rare Asian languages that are written with the Latin alphabet. But Vietnamese texts don’t compare to French, English or Spanish texts. Indeed, many marks, points, or signs are added to characters to transcribe all the sounds of this language or to indicate variations in speech. They are called diacritics. There exist lots of them in Vietnamese, and not only are they added atop characters, but also aside or below them, and several can be used at once. This way, an array of combinations are possible.
Often with handwriting, diacritics are not placed exactly where they should, and their sizes and shapes are distorted. It makes it difficult for recognition systems to accurately isolate segments and decipher meaning. Even more so when there are complex combinations, like in Vietnamese.
An error rate of only 3%
Hence, online recognition systems must be extremely efficient to recognize Vietnamese handwritten words, lines and paragraphs, with an error rate of less than 10%. In this competition, we owe our victory to our existing systems and to the new algorithms our researchers created for this specific challenge. Some of these algorithms will serve as material to keep maintaining the company’s technology at a cutting-edge level.
At MyScript, we have leading experts from all around the world in artificial intelligence, pattern recognition and linguistics. We power solutions for education, enterprise, automotive and end users in a large scope of languages: currently more than 65.
We offer the best handwriting recognition systems but most importantly, we took the power of digital ink to a whole new level: we created interactive ink. It is the only ink of its kind. It allows input cleanness, flexible editing, dynamic space management, machine-understandable and fully exploitable handwritten content, as well as operating system portability. In other words, naturally handwrite your content and edit it as if it had been written with a keyboard, for more productivity.
MyScript is ready for ICDAR 2019. Be prepared for our next incredible outcomes!