Tamil handwritten character recognition tested for robustness to image corruptions
uTHCD-C: A Corruption-Robustness Benchmark for Tamil Handwritten Character Recognition
Computer Vision and Pattern Recognition
Summary
Recognizing handwritten Tamil letters by computers usually works well with clean images, but real-life images can be messy and damaged. The authors created a new test called uTHCD-C that adds different types of image problems to see how well recognition systems handle them. They tested nine computer models and found that some are more resistant to these image problems than others. This helps set standards for making Tamil handwriting recognition more reliable in everyday conditions.
What this means in practice
- •For software developers: Build more reliable Tamil handwriting recognition apps that maintain accuracy despite image quality issues using standardized robustness benchmarks.
- •For mobile device engineers: Design efficient recognition systems that balance speed and corruption robustness for Tamil input on resource-constrained devices.
Authors
Preethi S, Meeradevi T, Arun Prakash A
Abstract
Evaluation of Tamil handwritten character recognition (HCR) has largely focused on clean test images, and a standardized multi-corruption robustness benchmark for isolated Tamil HCR has not been established. We introduce uTHCD-C, a corruption-robustness benchmark based on the unconstrained Tamil Handwritten Character Database (uTHCD). The benchmark defines ten corruption types, including a handwriting-specific stroke-thinning corruption, at five severity levels. We evaluate nine convolutional and transformer-based architectures using five-fold stratified cross-validation and writer-independent testing. Robustness is quantified using mean Corruption Error (mCE), normalized against a VGG16-BN reference. We additionally compare parameter count, computational complexity, and inference throughput to characterize the computational requirements of the evaluated architectures. The results establish comparative robustness baselines for Tamil HCR, revealing substantial differences across architectures: ConvNeXt and Swin achieve the lowest mCE values (0.75 and 0.83, respectively), whereas ShuffleNet and MnasNet exhibit higher values (1.55 and 1.85, respectively). The accompanying efficiency measurements characterize the trade-offs between corruption robustness and computational requirements. By providing standardized corruption configurations, evaluation protocols, and comparative results, uTHCD-C establishes a reproducible basis for future research on robust and computationally efficient Tamil HCR.