Image Description

Martin Scheid


Documentation of the Software Campus Project INDIGO

Design Canvas

Justificationary Knowledge

For the design of business processes the semi-formal modeling language "event-driven process chain" (hereinafter: EPK) and "Business Process Model and Notation" (hereinafter: BPMN) have become widely accepted in the business world. There, time-based and factual logical sequences are presented in the form of models. These models are used to document the processes and serve as a basis for further GPM measures. However, there are still a large number of modeling conventions for process design, which are strongly adapted to the needs of individual companies. In workshops for the collection of existing ways, processes are developed with the help of various tools in teams. Examples of utilities used include magnets in appropriate shapes, templates and classic flip charts. Depending on the size of the company, a large number of processes are often changed, which have to be digitized at great expense.


There are no appropriate possibilities to automatically convert sketched processes on flip charts or whiteboards into machine interpretable formats for further usage in information systems.


The primary goal is to automate or support the laborious manual and error-prone transfer process.


A software tool is created which accepts input in different image formats, analyses information about shape, structure and content of the business process and returns it in an adequate format.


A lot of frameworks focus on the analysis of image files. One prominent example is openCV which provides a lot of different algorithms and focuses on image processing. On the other hand the technologies around neural networks picked up speed and a lot of applications in the area of image analysis especially classification came up.

Design Process

In this project the DSR process according to Kuechler & Vaishnavi will be followed as an example for the design-process-templates which are available in MDP. This Process consists of five steps, starting with the "Awareness of problem"-formulation (I). On this the "Suggestion"-Phase starts (II), followed by the "Development"-phase (III), the "Evaluation"-Phase (IV) and the "Conclusion"-Phase (V). Whenever a phase is skipped or adapted, a statement is inserted why this phase is skipped.


How good are the trained ML-Models to identify a handwritten model? How good is the user experience when using this software?


INDIGO software tool for the classification of handwritten process graphs.

Design Knowledge

Instead of the staged algorithms known from state-of-the-art literature convolutional neural networks for picture semantic segmentation are used for the prototype.

Iteration 1: Project plan

Until the submission of the project idea, a project outline must be developed and specified.


Development of a proposal for the first project idea

Technical concept

Creation of the technical concept including the picture to server connection, so the image data can be read and analysed.

Scenario enhancement

To substantiate the benefit of the idea, an application scenario was created. This scenario is also utilized to improve comprehensibility.

Consolidation of the overall concept

Combines the application scenario with the individual aspects of the technical concept


This phase results in an initial proposal for a project outline.

Merging the contents

Merging the contents of the previous phase.

Preparation of the work plan
Preparation of the exploitation plan
Iteration 2: Project plan

Revision of the project description

Collection of the given feedback

The feedback of the supervisors is collected and merged so that I simplify the processing.

Processing of the given feedback

Not everything can be included 1:1 in the sketch Some things still have to be adjusted or deleted manually

Revision of the project outline

With the help of the feedback the project outline can now be revised

Coordination of the cost plan with the controlling department
Merge of all contents
Finalisation of the project outline
Iteration 3: Development of the first prototypes

Implementation using OpenCV

  • test
Awareness of Problem
Definition of subtasks
Literature Review
Definition of the first prototype based on previous literature reviews
Development of the first prototype
Creation of demo data
Evaluation using metrics
Evaluation through expert interviews
Summary of the results to date and identification of improvement potentials
Iteration 4: Development of the 2nd prototype

Implementation using Deep Learning techniques

Awareness of Problem
Literature Review

The emergence of new technologies, in particular deep learning, should drive the prototype in this direction.

Conversion of the prototype to Deep Learning

Based on the good results achieved so far with Deep Learning, the INDIGO prototype is also to be improved through the use of Deep Learning.

Implementation of Convolutional Neural Networks
Implementation of the Deep Learning prototype

Convolutional Neural Networks are used which are trained with the existing data.

Evaluation using metrics
Evaluation through expert interviews
Merging the results
Iteration 5: Further development of the prototype

The Deep Learning based prototype has achieved good results, but will need further developed to refine it.

Proposals for changing the network topology
Suggestions for changing the adjustable values
Adaptation of the network topology
Adaptation of variable values
Iteration via various settings
Examination of the metrics of all prototype test runs
Evaluation through expert interviews
Summary and evaluation of the tests

The project idea results from a master thesis where the topic was captured.

Feedback from supervisors

Feedback from the Software AG and the DFKI

Application for the Software Campus
Submission of the project outline

Submission to DLR.

Final feedback and release of the project outline

Final feedback and release of the project outline from the Software AG and the DFKI.

© 2024 MyResearchProcess. All Rights Reserved.