Automated Content Processing Platform through Deep Learning (ART/264CP)

Automated Content Processing Platform through Deep Learning (ART/264CP)

Automated Content Processing Platform through Deep Learning (ART/264CP)
01 / 11 / 2018 - 30 / 04 / 2020

Dr Arvin Wai-kai TANG

1. Image-Based Comparison Algorithm (IBCA) 1.A. Image pre-processing and matching algorithms for handling variations as: scale, rotation, translation, watermark, noise etc. 1.B. A Training platform framework and algorithms 1.C. A reference DL model for handling 20 pre-defined common traveling lost items 2. PCL Linguistic Extraction 2.A. Region-based Data Classification 2.A.a. A training platform and model architecture design 2.A.b. A reference DL model to perform unstructured 2D data segmentation and classification 2.B. Icon Classification 2.B.a. A training platform for icon classification 2.B.b. A reference DL model to classify the 50 Hong Kong widely used company/organization icons 2.C. Automobile Content Extraction 2.C.a. A training platform and model architecture design 2.C.b. A reference DL model and computer vision algorithms extract contents from images and express them in PCL linguistics for Automobile Insurance Claims 2.D. PCL Linguistic Expression 2.D.a. A training platform for PCL linguistic expression 2.D.b. A linguistics algorithm to refine textual data in PCL for insurance claim statements 3. Analytics and Modularized Rule Sets -- for users to define criteria based on the extracted contents 4. User manual, installation and maintenance documents for CS 1.

The Hong Kong Federation of Insurers (TL) [Sponsor]
The Hong Kong Federation of Insurers(CS) [Sponsor]

Content processing is essential to current modern financial activities and communication. The speed of creating is unbelievably fast and growth in the amount of 2D unstructured data is huge. To process and understand the 2D unstructured data is becoming a big headache to many big organizations.

This project aims to provide an Automated Content Processing Platform (ACPP) based on deep learning (DL) technology on analyzing unstructured 2D data for comparison:
- Understand contents from 2D data;
- Represent images in concise linguistics expression;
- A platform can be tailor-made for users' analytics/decision making based on the contents.

This is an integration of content extractions from unstructured data such as textual, photographic and pictorial images. A hierarchical structure is used to express the extracted contents in logical linguistics expression and human understanding. A set of modularized analytics tools is developed for users to make their own decisions on the linguistics expressions of unstructured 2D data, such as modularised comparison, cross validation and self-verification tools across single or multiple unstructured data. Image-Based Comparison Algorithms (IBCA) are also developed to compare image in pixel details. ACPP is applicable on any Enterprise Content Management (ECM), e.g., the Hong Kong Federation of Insurers (HKFI) will integrate it in their centralised database. The deliverables to HKFI contain two modules, IBCA for comparison on lost item images from travel insurance claims and Precise-Concise-Logical expression (PCL) for comparison on automobile damage claims.