If you have already uploaded your document to Transkribus and now want start working on it, the next step is the Layout Analysis, because in order to transcribe it, the image needs to be divided into text regions and lines. You will see, that it can be done with a few clicks and in most cases won’t require manual corrections. In the following video you will learn how to use the layout analysis feature.
In Transkribus Lite the layout analysis will run automatically, when you start a text recognition job.
Running the Layout Analysis
To arrive at the layout analysis tool, click on the “Tools” tab. The layout analysis section you will find right on top. You just need to select the pages you would like to run the job on and then hit “Run”. You can check the progress with the jobs-button. In the description-column it is indicated on which page the tool is currently working.
As soon as the layout analysis is finished the text regions (green) and the base lines (blue) will be shown in the image. To check the results, it is useful to activate the segmentation view at the viewing profiles. This way the text editor is hidden, and we have more space for the image to be shown. The text region covers the text. With more complex layouts, there can also be more than one text region on one page.
The baseline should run along the bottom of the text line, the letters should sit on it and the descender go below. If you do changes on lines, it is important to always do it on the baselines. This is important to know because, for every line in your document there is also a line region in the background. You can have a look at those by showing them with the item visibility button. These line regions must not be changed, they will be adapted automatically, when you change something at the baseline. There will be a pop-up asking you, if you would like to change the parent line as well, please confirm this.
Normally the layout analysis works with a high degree of accuracy. If changes should be needed, you can use the canvas menu left of the image to do so.
If there are sections in the document that do not need to be part of the training data you can simply delete the baseline, this way it will not be considered, for the model training. Click on it to select and then use the delete button in the canvas menu.
You can customize base lines and text regions by dragging these points or also split them with one of the split buttons.
With this button you will arrive at the selection mode again.
Two text regions or base lines, which are supposed to be one can be merged by selecting both, by holding control on your keyboard and then using the merge button.
Finished Layout Analysis
As soon as the layout analysis is finished you are ready to start transcribing in order to create training data. For this you will need to activate the transcription view again in order to see the text editor.
For the manual transcription please transcribe line per line, pager after page as close as possible to the original text.
As soon as you will have about 50 pages of manual transcription, we call it Ground Truth, you can start with the training of a model. If you work with printed material or use a base model, which fits to the writing in your document, the amount of training data can be even less.