Overview / Usage

Document image binarization, especially old handwritten documents, is a very important yet challenging task. There are various bottlenecks for binarizing historical documents due to different types of degradations present simultaneously such as back impression, ink bleed through, faded colours, and wear and tear of the writing media. We consider these degradations as various types of noise in the document image. Here we have proposed a 2D morphological network

which consists of basic morphological operation like dilation and erosion to perform our targeted task. The network also

includes linear combination of output from dilation and erosion operations. The aforementioned 2D morphological network is applied for image binarization, where the structuring elements (SEs) and the weights of the linear combination layer are learned through back-propagation. The proposed network has been evaluated on DIBCO 2017 and H-DIBCO 2018 and ISI-Letter dataset. Our results show more convincing as compared to the results of other state-of-the-art methods. Though the network is developed for old handwritten documents, it may be tuned to work for old printed documents also. The source code can be found here https://github.com/ranjanZ/ICDARBinarization

Methodology / Approach

1. We have defined 2D morphological Network for image processing tasks and utilize this network for binarization of old (handwritten) documents [see Figure 1]. SEs are learned using by back propagation method by minimizing a simple loss function.
2. We have made use of universal approximation power of Dense Morphological Network and proposed a morphological block consisting of dilation and erosion layers, and also a linear combination layer
3. The proposed 2D Morphological Network is built by concatenating these morphological blocks, which are capable to learn the spatial features from the input image.

Technologies Used

1. Intel AI dev Cloud

2. Intel _Xeon _ processor

Repository

https://github.com/ranjanZ/ICDAR_Binarization/blob/master/ICDAR_2019.pdf

Other links

• CODE (https://github.com/ranjanZ/ICDAR_Binarization)