Acquiring an OMR (Optical Mark Recognition) reading equipment can be impracticable to some companies or individuals due to its costs. Computational software solutions can be more attractive, but they require specific page format or page layout, such as specific marks to be used when recognizing any OMR document. In this paper, we propose a way to treat skew, translation, scale and alignment using a base document as reference due to the intrinsic characteristics of the problem. Key points are found by a pattern matching algorithm and used for the document image transformation. The method does not require specific layout, needing less formatting, allowing non-experts to create the form using ordinary software and scanners. Two experiments were executed: one with 40 images distorted randomly from a document clipping, and the second one with 1034 images of real student tests. Both experiments reached high overall accuracy.
20 January 2026 in Journal of the Brazilian Computer Society
To address sample scarcity, a common challenge in medical imaging, we investigate the Reptile meta-learning algorithm for few-shot disease classification in fundus images. In a setup with eight training classes and four testing classes, we investigate different architectures, training strategies, varying N-way K-shot configurations, and different data augmentation techniques. Our results show that Reptile outperforms standard transfer learning approaches in several settings and that, when combined with data augmentation, especially during evaluation, correct predictions tend to receive higher confidence scores. The quantitative results are further supported by an analysis of prediction confidence levels and activation-map visualizations.