Text recognition accuracy assessment

How the BIQE OCR Server achieves OCR accuracy

Recognition accuracy assessment as a tool for obtaining the best result

Text recognition accuracy assessment

Information recognition and extraction are essential functions of the BIQE OCR server. In order to provide our customers with the best recognition quality, we use one of the smartest recognition engines in the world, ABBYY FineReader Engine. (Upon request, we can use any other recognition engine the client wants.)

Our clients usually process large (and even huge) file arrays (scans, images, pdf). Therefore, it is valuable and important for our clients to obtain summary information on file processing to assess the quality of processing and detect and correct possible problems.

For example, low-quality scans, blank pages, upside-down pages, images containing a lot of garbage, etc. The BIQE OCR server helps our customers conveniently analyze and improve the processing results.

BIQE OCR server displays valuable summary information about the results of processing each page. This information is available in tabular form. A separate column in this table shows each page’s estimated value of recognition quality/accuracy. The operator can sort the table by this column and see sorted blank pages or pages with a low recognition quality score. The operator easily finds problematic pages. Then, he can correct them in the general package of files.

Thus, the BIQE OCR server helps our clients conveniently control the quality of processing and obtain the best target result.

Calculating Recognition Accuracy Score

The assessment of recognition accuracy is not just a criterion, but a pivotal factor that determines the quality of file processing. It’s the cornerstone on which the analysis of the target result and the final quality of file processing are based. Therefore, the recognition accuracy estimate must be calculated with utmost precision and reliability.

Typically, recognition engines evaluate the confidence of recognising words and individual characters, including ABBYY. The recognition confidence level depends, first of all, on the quality of the scan and some other factors (for example, semantic analysis).

The confidence level basically shows how well the recognition engine “recognised” the character or word. This criterion is useful for assessing how well the recognition engine is trained to recognise a particular text and font. However, it will not allow you to evaluate the accuracy of recognition.

For example, when recognising a low-quality scan, the recognition result may be 100% correct and accurate, but the level of confidence that the engine provides can be very low. Therefore, this criterion is not an assessment of recognition accuracy.

Remember, the BIQE OCR server calculates the recognition accuracy score using a complex algorithm that explicitly considers the confidence level indicator and other criteria. This underscores the importance of recognition accuracy in our work.

Would you like to learn more?
Please contact us, we are happy to help you!
info@biqe.biz

Postal address
Meerweg 17
8313 AK Rutten
Netherlands