Archival Document Processing using Cognitive Computing
Patel, Himaniben Pareshkumar
This item will be available on: 2020-05-01
The world, as we know it, is constructed in the form of knowledge. Our ancestors have passed their experiences to the next generation over time using handwritten documents. Although these old manuscripts are still available however, to disseminate that information to everyone, they must be converted into digital form. In the 21st century, the computers are becoming faster than ever before, thanks to the advancement of the fields of machine learning, deep learning, big data, cognitive computing and etc. A relationship between data may be found, which may, in turn, solves most of the problems. Cognitive computing can be used to deal with a vast amount of data to discovers hidden patterns or insights. Although research has explored many diverse, specific fields of application for cognitive computing, a comprehensive overview of the concept and its use is severely lacking. By leveraging the abilities of cognitive computing, text may be extracted from the handwritten documents in the form of images. The first part of the thesis focuses on the literature review of research papers related to applications of cognitive computing, collected from IEEE, ACM, and Springer databases. Currently, two companies provide cognitive computing services related to handwritten text recognition, Microsoft Azure's Computer Vision and Google Cloud's Vision AI. The second part focuses on conducting a performance analysis between these services based on some pre-defined criteria, where Microsoft Azure's Computer Vision service performed better overall for cursive English. Transkribus is a platform for automated recognition and transcription of archival documents, which uses a deep learning model to recognize text from an image. The third part focuses on analyzing the effectiveness of Microsoft Azure's Computer Vision service, by conducting performance analysis with Transkribus where images (collected from the Library of Congress with their transcribed text) were submitted. The results showed that Microsoft Azure's Computer vision service performed better compared to Transkribus. The last part focuses on increasing the accuracy of the Microsoft Azure's Computer Vision service by improving the quality of images. Various image pre-processing techniques were analyzed and applied to the dataset. Both improved and un-improved images were given as input to Microsoft Azure's Computer Vision service, and their results were evaluated, which showed that Microsoft Azure's Computer Vision's accuracy could increase for some images by improving the quality of the image.
Patel, Himaniben Pareshkumar. (July 2019). Archival Document Processing using Cognitive Computing (Master's Thesis, East Carolina University). Retrieved from the Scholarship. (http://hdl.handle.net/10342/7489.)
Patel, Himaniben Pareshkumar. Archival Document Processing using Cognitive Computing. Master's Thesis. East Carolina University, July 2019. The Scholarship. http://hdl.handle.net/10342/7489. September 20, 2019.
Patel, Himaniben Pareshkumar, “Archival Document Processing using Cognitive Computing” (Master's Thesis., East Carolina University, July 2019).
Patel, Himaniben Pareshkumar. Archival Document Processing using Cognitive Computing [Master's Thesis]. Greenville, NC: East Carolina University; July 2019.
East Carolina University