Documents in the form of scanned, copied, and photographed image files, ID cards, and various certificates/applications, etc., are text-recognized (OCR) to extract personal information in the image and mask the information so that it is not identified within the image.
Classification of documents containing personal information
· Personal information filtering awareness · Personal information keyword recognition · Recognition of specific certificate titles · Automatic document classification · Classification log inquiry
Character recognition (OCR) personal information masking
· Image character recognition · Text and attribute extraction · Recognizing personal information patterns · Personal data logging · Create masking information of personal information
Accurate classification by checking whether personal information is included in each step
(Step 1) Filter resident number to check whether resident number is included or not (Step 2) Classification through recognition of keywords and certificate titles related to personal information when resident number is not detected [keyword] : Resident number, resident registration number [Certificate recognition] : Resident registration card, driver's license, passport, a copy of resident registration, seal certificate, family relation certificate, basic certificate, etc.
Diversity of document recognition
· Formal/atypical documents · Image file and PDF file · Files containing multiple pages such as TIFF and PDF
Masking the resident number and fingerprint information by text recognition
Performing logging of personal and document information after masking the resident number from car registration certificate image.
OCR server OS
· MS Windows 7, 8, 10 (64-bit only) · Windows Server 2012, 2016, 2019 (64-bit only) · Linux (64-bit only)
OCR Server specification
· Intel Xeon 2.3Ghz 4-Core or higher · Memory 8GB or higher · HDD 300GB or higher
It is a solution that identifies fingerprints from scanned images or photographed photos, and masks the location of fingerprints, and provides privacy by de-identifying the fingerprint.
Fingerprint identification
· Fingerprint identification through similarity check using deep learning technology · Extraction of location and size of identified fingerprint
Fingerprint masking
· Fingerprint de-identification processing through block masking according to the fingerprint location and size
Support file format
· JPG, BMP, PNG, GIF, TIF, PDF
- Fingerprint location detection using deep learning technology from images of low-resolution ID cards or unstructured documents
- Unlike biometric fingerprint (directly recognizing the integrity of the fingerprint with an optical camera) a deep learning technique calculates the similarity threshold using deep learning matching.
Fingerprint recognition and masking process speed
Test environment
Hardware · Intel Core i7-4 CPU, 8GB memory · GPU : NVIDIA GeForce GTX1080 Ti D5X 11GB software · operating system: Windows 10 64 bit Test image · Multi Tiff 10 pages · Capacity: 1 MB (average 100 Kbyte) · DPI: 200 dpi
· Uses dedicated GPU(Graphics Processing Unit) - mounted hardware for fast deep learning image process · Hardware can be configured as server/workstation or high-end PC
Test speed results
Total Time(Single Thread, Single Processing) · 8 seconds (average 0.8 seconds per page/total 10 pages) Time required for each step
OS
· MS Windows 7, 8, 10 (64-bit only) · Windows Server 2012, 2016, 2019 (64-bit only) · Linux (64-bit only)
Server specification
CPU version · Intel Core i7/i9 (4 gen. or higher) or AMD Ryzen · Memory 8 GB or higher · HDD 300 GB or higher GPU version · Intel Xeon 2.3Ghz 4-Core or higher · Memory 16 GB or higher · HDD 300 GB or higher ※ GPU : NVIDIA Quadro-P4000 or higher (GPU when configuring high-end PC Workstation: Geforce GTX or RTX series memory 6GB or more)