![]() ![]() We paid considerable attention to this user pain point and wanted to help our users by giving them the tools to obtain a perfectly flat electronic document with ease. If there is curvature in the image, the correcting effect is still not satisfactory. This helps users in certain cases, but it only solves the problem of perspective correction. So, what could we do in this situation? Some tools on the market can perform perspective correction on an image from a distorted perspective like below: Presumably, you cannot always carry a scanner with you. (Note that the texts are not comfortable to read, given the curvature of the page) Because the book is fairly large in size, the particular page cannot be fully flattened, which makes it a challenge to take a quality picture for reading later. What is document image dewarping?When reading a book, many of our users run into issues when they want to then extract an excerpt for personal notes. Today, we will focus on just one use case - document image dewarping. Note that all these functionalities happen behind the scenes so we encourage readers to download our app on Google Play and Apple App Store, and give them a try. Layout analysis: when converting images to documents, we use both TensorFlow and scikit-learn to implement layout analysis these frameworks considerably reduced our engineering efforts. ![]() Offline OCR: we use TensorFlow Lite to run OCR models on mobile devices and localize texts quickly and accurately for users who are sensitive to data security.Online OCR: we deploy TensorFlow models on our servers to perform rotation correction, text line detection, and more on user documents to save editing time.Automatic document category classification: we use TensorFlow Lite to implement a machine learning model that is able to automatically classify documents based on document images, and provide corresponding functionalities based on the category.Document edge detection: the CNN-based rectangle detection network devised at WPS can quickly identify the edges of target document, solely on-device.We will elaborate on this in the next section. Document dewarping: automatically flatten a warped document image.Using TensorFlow, we can help our users to process their documents in a more intelligent manner in a number of use cases, including: We strive to provide the best tools and features to our users so that they can get their document processing done efficiently. Machine learning use cases at KingSoft WPS Office from Kingsoft is a productivity tool that serves 150M+ users globally. A guest post by Longfei Xiong, Cheng Du, Ronghua Chen, Hui Zheng and Xuhua Hu from WPS AI Engineering Team
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |