Whether you need an invoice annotation for a quick expense report, professional accounting project, or extensive financial analyses, this particular type of text annotation task can be complex and time-consuming. With all the available photo and scanning technologies, it has become much easier to collect receipt scans, however, most of the annotation tools don't seem to catch up with such a speed. Without the possibility of labeling your data as effectively as possible, the time gained on data collection will be inevitably lost on document annotation.
How to use SuperAnnotate to speedup your receipt annotation task.
Annotating text in invoices or in financial documents is a much different task than a common object detection or a semantic segmentation. In the case of text annotation, the use of computer vision is directed more towards tags and labels and attributes rather than classes. Ideally, the annotator should be able to switch between free-text labels, predefined attributes, and classes seamlessly in order to save time. On the other hand, text annotation unlike other types of image annotation is based on groups: groups of words or notions that are in the same category. This means the annotator should also be able to group multiple objects together almost at the speed of a reflex. Conclusively, the operations needed for annotating text images are the following:
1. Automating bounding boxes via predictions
Assign each word or number a bounding box to define its location relative to the image space․ This task can be automated by using the smart predictions integrated within the platform. In most of the text-based images, our prediction algorithm is able to receive 97% mAP score and can recognize over 50 different languages.
2. Text assignment
Assign each box an appropriate text. The text assignment can be also automated along with predicting the location of the bounding boxes. After the prediction is finished the annotator can now fix the errors left out by the system. A set of free text and predefined attributes, along with classes give you a 3 level flexibility to speed up the small portion of manual work left after running the smart predictions.
3. Object grouping
Categorize the boxes according to groups and subgroups that might be needed for a financial data annotation. The grouping and ungrouping of already labeled boxes are instant within the platform.
By combining the abovementioned techniques, the platform offers a complete toolset, designed to accommodate the intrinsic specificities of text annotation. A prediction API capable of recognition over 50 languages and a flexible vector toolset allow the user to automate the time-consuming task of text image annotation on all operational levels.