Seamless OCR

Shriram KV

Shriram KV

Bengaluru, Karnataka

OBJECTIVE: Seamless Optical Character Recognition Bundled within an Intuitive and Easy to Use Android Dashboard. Proposed Solution: We’ve built a system that recognizes characters (handwritten and printed) with high accuracy (99% ). The algorithm is deployed on the cloud. ...learn more

Project status: Published/In Market

HPC, Artificial Intelligence, Cloud

Intel Technologies
DevCloud, Intel NUC

Overview / Usage

Proposed Solution: * We’ve built a system that recognizes characters (handwritten and printed) with high accuracy (99% ). * The algorithm is deployed on the cloud and is accessed via an API. * The user, provided with an android interface, has to upload the required image. The app routes the image to the cloud where the algorithm processes it and results are obtained in real-time.

Salient Features : * Highly Robust, can process all inputs with relative ease. * The entire ecosystem is Plug-and-Play, hence the user experience and ease of adoption are quite high. * Highly accurate Diagnosis. * Scalable as the entire system runs on the cloud, and results are obtained in a quick manner.

Methodology / Approach

How does it work? This product primarily focuses on handwriting recognition on the exact sciences application form, we take a picture and send it to our handwriting recognition module on the AWS server. Firstly localization will be done in order to identify where the handwriting may be present and the coordinates would be noted, as a next step, each and every content in the bounding boxes will be sent into our deep learning model which is capable of identifying different handwriting styles. It essentially consists of a feature extractor of 5 CNN layers added to 2 bi-directional LSTM having CTC loss function. All the content present in each and every bounding box will be identified precisely and the entire thing, the different fields in the application form and corresponding identified handwriting text will be stored in a nonrelational database. We provide 2 APIs one for upload and one for accessing the stored content in the DB. On uploading, the user can see what is the identified text immediately and redo if there is any rectification. After successful upload, everything will be stored in DB and the entire DB will be accessible by the admin

Technologies Used

AI, OCR

Collaborators

There are no people to show.

Comments (0)