In the context End-To-End Measure for Text's crucial to evaluate the performance of text recognition and text line detection engines before comparing systems and configurations. Both jobs have established metrics that can be used independently.
To evaluate the effectiveness of a system that combines text line detection with text recognition, however, there is no sophisticated system in place.
A well-known methodology that is occasionally applied in this context is the F-measure on word level. However, it can produce misleading findings because it does not account for the alignment of the hypothesis and the ground truth text. There is a considerable demand for such a metric since users of automatic information retrieval pipelines in the context of text recognition are primarily interested in the end-to-end performance of a given system.
As a result, we offer a metric to assess a text recognition system's overall quality. The widely accepted character error rate serves as the foundation for this measurement and is only applicable to aligned hypothesis and ground truth texts in its original form. The suggested measure is adaptable in that it may be set up to punish various reading orders between the hypothesis and the actual data as well as take into consideration the text lines' geometric positions. It also has the ability to disregard the over- and under-segmentation of text lines. It is possible to obtain a measure that best suits its own demands using these factors.
To know more about Hypothesis visit: In the
brainly.com/question/28104822
#SPJ4
