A lot of work has been done in the field of document image binarization. Is it a closed research field or still an interesting challenge area? I am new in the field, and just curious.
First, in science, a field is rarely closed, sometimes asleep only. Resistance to low-contrast, real-time, badly scanned, composite documents/writers or from aging medium seem to remain challenges, in similarity with other digital data: robustness, speed, poor acquisition, source separation or error concealment (the same topics, with generic names) are common for other data source.
As many data fields, this domain is being pervaded by (deep) learning techniques threatening the old handcraft (eg Learning Document Image Binarization from Data, Yue Wu et al., 2015).
DIBCO 2017, the ICDAR 2017 Document Image Binarization COmpetition, is still on, a good sign altogether, and a source of challenges.7
As a personal note, I have been very interested in the analysis of old written documents, and their authentification, based on:
- binarization from alternative light sources (multispectral, X-ray, ultraviolet), especially to detect previously scrapped, washed off or erased writings, like from palimpsests (Binarization of MultiSpectral Document Images, 2015)
- accurate analysis of non-written elements like the texture of paper fabric