13.4.2.3 A.I. Image Recognition Parsing

Parser recognizes .jpg, .jpeg, .png & PDF images and can automatically detect the shapes and positions of BPMN flow objects, text, and transitions, roles and lanes. This allows organizations to upload their existing process map images and have them automatically exported to EPC without having to recreate them from scratch in the web app.

How to get started:

1. Upload your process images to Parser. You can upload more than one at a time. This operation may take a couple of seconds.

2. Click on to continue to the next page.

3. In the next page, choose or create a template. This step does not affect the parsing. Click on to continue.

4. Unlike document tagging, image tagging is done automatically and CANNOT BE MODIFIED.

5. If no process name is detected by the Parser in the image, the process will have the name of the image.

6. After analyzing your image, click on . This operation can take a couple of moments.

7. Notice that the “Export” button has been greyed out. You can only export an image once.

8. If you uploaded multiple images, navigate to the “Current Source(s)” drop-down to switch the image you are viewing. Repeat the previous step of exporting. Each image needs to be exported.

9. Navigate to your EPC webapp and refresh the page.

10. The process will be displayed in your tree. The map layout and flow will be determined by EPC’s auto-layout rules, so it may not be identical to your original image.

Current Limitations and Rules

• If you see that an image has minor parsing errors, which can happen with any AI tool, you can easily modify it in the EPC graph edit.
• We recommend you use the “Create new objects” option in the Export Settings when parsing an image.
• If your flow objects are not named in the image, they will be given a default name that will be editable in the EPC webapp. If there are multiple flow objects without a name, they will be numbered (e.g. gateway1, gateway2, etc).
• Parser is currently trained primarily to recognize horizontal process map images. While vertical images can also be parsed, they might result in a slightly higher error rate.