OCR Studio cuts ID recognition compute by 32%

OCR Studio has unveiled an ultra-lightweight neural network that can recognize IDs with 32 percent fewer operations than current advanced models.

The developer of ID scanning solutions has made the new AI technology available on mobile phones, tablets and desktops. The technology is already used by the company’s clients in banking, retail and healthcare. The invention was presented at the 18th International Conference on Machine Vision (ICMV 2025) in Paris.

Industries such as banking, logistics and border control widely employ automatic identity document recognition while today’s ubiquitous mobile platforms require the relevant technologies to be performant even on budget devices with more limited processing power.

Recognition models need a broad receptive field to maintain high computational accuracy and effectively handle challenges such as projective distortions, noise, motion blur and poor lighting in document images. However, models with such expansive receptive fields often require substantial computational resources, making them too heavy for real-time document analysis on mobile devices creating a bottleneck in often fast-paced environments.

OCR Studio tackles this challenge by incorporating a block of learnable Infinite Impulse Response (IIR) filters into its solution. Unlike most advanced methods, these filters dynamically adjust the receptive field by factoring in both current inputs and previous outputs. This innovative approach was tested on the MIDV-500 and MIDV-2019 datasets, where it achieved a three percent reduction in computational load for ID recognition compared to leading alternatives.

The technology is now embedded in the latest version of OCR ID-scan, which is actively used across government, fintech, human resources,and other sectors to streamline document-processing workflows.

The OCR Studio team is set to appear at an international scientific conference, a first for the team, with the publication of a paper on its IIRDoc-Net technology in the electronic proceedings of the International Society for Optics and Photonics (SPIE). The paper will be presented at the International Conference on Machine Vision (ICMV), where proceedings are indexed in leading research databases including the Web of Science Conference Proceedings Citation Index-Science, Scopus, and Ei Compendex.

Article Topics

document reader  |  ID verification  |  identity verification  |  OCR  |  OCR Studio  |  research and development