Map Simplification

Digitization of old records is a crucial task for organizations worldwide, as it not only preserves valuable historical information but also enhances accessibility and usability. By converting physical documents into digital formats, organizations can better understand historical changes, improve record-keeping practices, and facilitate the upgrading of data infrastructure. This process is essential for maintaining the integrity of information while ensuring that it remains available for future generations. 

This model can generate a simplified version of scanned images, including topographical maps, floor plans, and cadastral maps. It helps to simplify hatch patterns, reduce salt-and-pepper noise, and effectively separate background from foreground elements. By extracting higher-level patterns from the input imagery, the model enhances the clarity and usability of the data. During the image simplification process, the input images are converted to black and white, which helps in emphasizing the essential features and structures within the maps. This conversion is beneficial as it streamlines the semi-automatic or manual digitization workflow.

Using the model
Follow the guide to use the model. Before using this model, ensure that the supported deep learning libraries are installed. For more details, check Deep Learning Libraries Installer for ArcGIS.

Fine-tuning the model
This model cannot be fine-tuned using ArcGIS tools.

Input
8-bit, 3-band image.

Output
A simplified, black & white variant of the input 3-band image.

Model architecture
This model is based on the MangaLineExtraction architecture by (Li et al., 2017).

Sample results
Here are a few results from the model.