English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Reading dirty license plates accurately in moving vehicles is challenging forautomatic license plate recognition systems. Moreover, license plates are oftenintentionally tampered with a malicious intent to avoid police apprehension.Usually, such groups and individuals know how to fool the existing recognitionsystems by making minor unnoticeable plate changes. Designing and developingdeep learning methods resilient to such real-world 'attack' practices remainsan active research problem. As a solution, this work develops a resilientmethod to recognize license plate characters. Extracting 1057 character imagesfrom 160 Nepalese vehicles, as the first step, we trained several standard deepconvolutional neural networks to obtain 99.5% character classificationaccuracy. On adversarial images generated to simulate malicious tampering,however, our model's accuracy dropped to 25%. Next, we enriched our dataset bygenerating and adding geometrically masked images, retrained our models, andinvestigated the models' predictions. The proposed approach of training withgenerated adversarial images helped our adversarial attack-aware license platecharacter recognition (AA-LPCR) model achieves an accuracy of 99.7%. Thisnear-perfect accuracy demonstrates that the proposed idea of random geometricmasking is highly effective for improving the accuracy of license platerecognition models. Furthermore, by performing interpretability studies tounderstand why our models work, we identify and highlight attack-prone regionsin the input character images. In sum, although Nepal's embossed license platedetection systems are vulnerable to malicious attacks, our findings suggestthat these systems can be upgraded to close to 100% resilience.

Adversarial sample generation and training using geometric masks for accurate and resilient license plate character recognition