History/Change Logs

Version 3.5

Released on 14 November 2022 Overall Accuracy = 98.13%

New Features

1. Our OCR now can support PNG, JPG, JPEG, HEIC, HEIF and PDF file extension for On-Cloud and On-Premise.

2. Our OCR now can support option for get the bounding box of fields (not be able to using auto rotation), get the image in base 64 format and get the original value from OCR engine in the JSON response.

3. Users now can select desirable text fields that they actually want to perform OCR (which make OCR process faster), by specify in "/config/config.yaml" (On-premise version) or specify as parameters in URL requests (On-Cloud version) which support following fields.

4. Users now can define maximum size of File requests, by specify in "/config/config.yaml" (On-premise version)

5. Included confidence score config, options config, fields config and system settings to config.yaml file (On-premise version)

Bugs Fixed

1. Fixed bug the back card side not found when not taken a photo at the angle perpendicular on the card.

Version 3.4

Released on 1 September 2021 Overall Accuracy = 93.25%

New Features

1. Users now can select desirable text fields that they actually want to perform OCR (which make OCR process faster), by specify in "/config/fields.txt" (On-premise version) or specify as parameters in URL requests (On-Cloud version) which support following fields.

Fields Name	Fields Description
id_number	13 Digits National ID Card Number
th_name	Thai person name
en_fname	English person first name
en_lname	English person last name
th_dob	Thai date of birth
en_dob	English date of birth
address	Thai Address
religion	Thai Religion
th_issue	ID Card's Issued Date in Thai Date Format (B.E.)
en_issue	ID Card's Issued Date in English Date Format (A.D.)
th_expire	ID Card's Expired Date in Thai Date Format (B.E.)
en_expire	ID Card's Expired Date in English Date Format (A.D.)

2. Every OCR result of a 13-digits "id number" field is now validated with the last digit checksum rules. If the validation failed, the error_message will notify the invalid result. 3. User now can config that the OCR should either accept the black/white picture, color picture and gray scale picture. 4. Our OCR now can support .heic and .heif file extension but limit only for On-Premise.

Bugs Fixed

1. Better Queue Management is implemented, less "Server is busy" 530 error message. 2. Abbreviation of April in Thai is wrongly misspell to "ม.ย.", We fixed it to "เม.ย.". 3. Fix OCR output that wrongly split two-character alphabets to two alphabets, such as "อ|๐|า" -> "อ|ำ" and "เ|เ|อ" -> "แ|อ". 4. Add more Thai name initials vocabulary (Thai military ranks). 5. Improve model by 0.5% (92.75% -> 93.25%)

Version 3.1-3.3

Internally released for closed-beta to increase the accuracy. Released during May-July 2021 Overall Accuracy = 92.75%

Version 3.0

Released on 14 February 2021. Overall Accuracy = 91.75%

Version 2.0

• Add Backcard OCR.

Released on 1 February 2020 Overall Accuracy = 85%

Version 1.0

Released on 1 November 2019 Overall Accuracy = 76%

• First initial version

• Support Only Front Card

****

---

---