This survey on text and no-text separation in images presents a quite complete review (list of references) of image document analysis, including printed and handwritten texts. The authors present tables comparing the performance of the methods found in the literature; however, the results are usually not directly comparable since they consider different datasets.

Although this paper is interesting and presents a very good review of many different methods, there are some drawbacks. First, the authors classify the different types of image documents into four classes, but mix both printed and handwritten texts in classes 2, 3, and 4. Handwritten text segmentation should be considered separately, since it is very different from printed text analysis. There are probably different and better handwritten document categorizations than the one presented here.

Next, although this paper is related to documents, one very important text and non-text separation problem is related to reading real-world text using cameras, that is, text detection and segmentation can be more complex than well-aligned/well-posed texts in documents. What about capturing text from a distorted view (camera not aligned with the text, text detection and segmentation from real-world camera images), which can be very helpful for people who rely on automated reading devices.

Finally, while the included tables present references and performance over different datasets, a table that summarizes the different features and approaches found in the literature would be more useful and interesting.