Handwriting Recognition- how efficient is it?

There are many types of programs that recognise handwriting but they are different from each other in terms of the technology they use, and also the type of data they can manage.

Handwriting recognition is a difficult task for a program as well as for people (sometimes we have difficulties recognising our own handwriting). The handwriting recognition technologies have been developing for the last 20 years but they haven’t reached a final stage yet.

One of the reasons why developing such a system takes so much time is the learning period. But, before the program can turn the handwriting into digital text, it needs to determine which specific characteristics of the letters (or other symbols) it needs to identify in order to recognise them. This action is performed by some classifiers, each of them searching for and analysing certain sequences of the characters (for example the round curve to the left of letter “d”). Then, the system needs to “learn” to recognise the multiple variations that can appear in handwriting and that can change the look of the letters, their angle, the spaces between them etc. Most of the times, the learning process is not an actual learning process in itself but it rather refers to the adding in the data base of new templates of how letters can look like.

There are many options of programs that can perform handwriting recognition, but they are not all the same. This is not only from the point of view of the technology they use, but also of the type of data they can manage. There are significant differences between the handwriting restricted to boxes (each character appears in a separate box, for example the 13 boxes used for CNP- personal identification number), capital letters and the normal handwriting. This is why, as far as the handwriting recognition is concerned, it is difficult to mention a certain level of accuracy, as it is done in the case of programs that recognise handwriting with capital letters. Accuracy depends on many factors such as: the legibility of the handwriting, how much it drifts away from the norm, if there are characters with unusual features and even the quality of the paper. If in the case of capital letters there is a limited, though high number of possible fonts, in this case the font can be slightly different from person to person.

Nevertheless, some handwriting recognition programs are quite performant and thus they are included in the gadgets that we use daily. Some tablets incorporate in their programs handwriting digitalising systems. Others, such as E-Diary, can be purchased separately. This software can recognise notes or more complex texts which were written on paper or straight on a tablet. But generally, some people prefer to write digitally because it is faster than handwriting. In other contexts though, these technologies can be extremely useful. For example, a researcher at MIT created new software that recognises handwriting in the students’ papers in math and science. Bases on this, students can get almost immediate feedback regarding their performance. This way, teachers don’t spend time correcting the papers but they can focus on providing feedback and on correcting the most common mistakes identified by the program.

These programs can also include another application namely the automated extraction of handwriting from structured or semi-structured documents. It is estimated that 80% of all the documents used in industry, business, trade, financial transactions and administration are semi-structured. Most documents are filled in by hand, which means that the data entry operators have to extract the information manually. By using handwriting recognition programs the extraction process is much more efficient. These programs will continue to require the assistance of the operators for checking, but the costs will be much lower compared to the traditional version.