TH22363Y - Optical Character Recognition (OCR) system powered by artificial intelligence. - Google Patents

Abstract

This invention relates to an optical character recognition (OCR) system from AppMAN. It assists in character recognition or character detection based on data fed to an artificial intelligence (AI) system. It helps in creating or reading data and can be applied or integrated into five workflow processes depending on the application. The process consists of: Inputting the document (e.g., national ID card, vehicle registration book, or other commonly used documents) into the system; Image Pre-processing/Transformation, which involves cropping the image to the original document's proportions; removing noise through image manipulation; and Text Detection, where the text is detected within the cropped image. This data is then fed to the AI system, which has been trained on image clarity and text blur to improve accuracy across all document types. Finally, the system performs text recognition and then synthesizes the information by matching the meanings within the input document. The final processing step (Post 15 processing) yields the result as fill text within each section of the document. This invention's system also includes a data training system and a data reading development system to ensure comprehensive and up-to-date information, along with a security system for the Appman optical character recognition (OCR) system.

Claims (4)

1. AppMAN's Optical Character Recognition (ACR) system assists in character recognition or character detection based on data fed into the artificial intelligence (AI) system. It helps in creating or reading data and can be applied to or integrated into various work processes depending on the application, such as document recognition, ID card recognition, and connecting to Know Your Customer (KMTC) forms of various organizations to improve data entry efficiency, speed, and accuracy, reducing human error. The product operates via an online cloud server and also on an on-premises server system, and can connect to websites. This system, connected via an API, can be integrated into web applications such as various web browsers. The operating principle involves Optical Character Recognition (OCR). This method consists of three steps: A. Inputting the document data that the artificial intelligence (AI) system wants to read. The company's OCR software is fed with source documents such as national ID cards, vehicle registration books, or other commonly used documents in the country or internationally. B. Cropping the image to the proportions of the source document (Image Pre-processing/Transformation), then reproducing the image to remove noise such as blurriness or blurred text. C. Text Detection in the cropped area, based on the data fed to the AI system. The AI system has been trained in aspects related to image clarity and text blurriness to increase accuracy in reading the data. All aspects of document data are processed as follows: D. Text Recognition: After process C, the Artificial Intelligence (AI) system will read the characters in the document data we entered into the system. E. Information Extraction: The AI system will match the meanings after process D. For example, the text in this field could be name, surname, national identification number, vehicle registration number, or other meanings in the document data we entered. F. Post-processing: This is the final processing step, and the result will be the text in each section (fill text) of the document data. 2. The AppMAN Optical Character Recognition System, as per claim 1, includes a data training system and the development of comprehensive and up-to-date data reading models. It begins with data diagnosis, resulting in fill text for each section of an online document (soft file). This is then passed to a five-step data training process. In this step, humans verify the results for accuracy (e.g., identifying '1' as '7', 'a' as 'ci', similar characters, similar alphabets), and whether the results can be used to develop a document model for improvement. If the results are deemed suitable for model development, humans will review and verify the document data, then process the data by dividing and describing the overall components of the document (e.g., name, surname, ID number, vehicle registration number, header, footer, or the entire document), clearly separating and describing them. Regarding accuracy and error, the process involves managing the file, creating a solution, and ensuring the document is 100% accurate. This is then submitted to software engineering for model development. After the data training phase, the model development phase begins, which involves inventing and synthesizing new document models to continuously improve and update the results. Once the model is developed, it is tested against the core principles of machine learning. If the test results are satisfactory, the model data is updated with multiple models and formats in the AI database. Examples include adapting the appearance of national ID cards to accommodate future laws, including ME-0, JC-3, non-religious identification cards, vehicle registration books, passports, or invoices and credit card documents to better support various document versions and formats in the future, enhancing the efficiency and capabilities of the AI and updated machine learning cores. If the test results are not sufficiently accurate, the document data is re-checked and verified, and then the data is managed, the file is processed, and the solution is created to ensure the text is correct. The document is completed, and new document models and formats are devised and synthesized to continuously improve the results. These are then tested against the core principles of machine learning. If successful, the new model data is updated. 3. AppMAN's Optical Character Recognition (ACR) system, as per claim 1 or 2, further includes a security system for the ACR's operation. When inputting data (input for each service) for the AI system to read, the authorization system will verify the username and the entered data. The system will check and confirm all evidence. If correct, authorization will be granted; otherwise, it will be denied. The authorization system will also record which user entered data and what data was entered. Once authorization is complete, the system will connect the data to a Virtual Private Cloud (VPC Link) and manage it via a load balancer to balance the incoming data, preventing system crashes and distributing data appropriately. This ensures optimal server performance and prevents server failures. The input data will then be coordinated with... The internal application interface includes a sub-data management system, an application interface manager, and 15 other systems unrelated to the core computer learning model or other applications. These systems are then ingress-facing, allowing data to enter the internal interface and be appropriately distributed by the load balancer. 4. AppMAN's optical character recognition system, as per claim 1, can read 20 documents including: Thai Citizen ID, Indonesian Citizen ID, mobile banking slip, truck/car/motorcycle registration, passport, receipt, invoice, lawsuit, certificate, credit card, or other documents as may be added in the future.