Adaptive fuzzy model for blur estimation on document images

In this paper, we propose a local blur estimation for document images captured by portable cameras. A novel blur pixel feature is extracted from pixel's properties in working zones to initialize a fuzzy clustering of blur and non-blur classes. At the final state of the process, a blur region is determined for each working zone. The blur score is given by the average of all membership values of pixels in the blur region. The quantitative evaluation on two real databases (DIQA and an industrial database) shows that our method achieves good results in comparison with recent methods estimated on these databases.