Akademik Veri Yönetim Sistemi
8th International Congress on Human-Computer Interaction, Optimization and Robotic Applications, ICHORA 2026, Ankara, Türkiye, 21 - 23 Mayıs 2026, (Tam Metin Bildiri)
This study presents a novel pipeline for both high- and low-level vision tasks by implementing four proposed models exploiting the evolution of the transformer-based models in the diagnostic labeling and restoration domains. Researchers curated a dataset of 500 degraded oral cancer images and 250 non-oral cancer samples incorporating SwinTransformer, VisionTransformer, Hybrid-I, and R-Vi-Swin-Net. Hybrid-I integrates SwinTransformer and VisionTransformer while the R-Vi-Swin-Net pipeline synthesizes VisionTransformer, SwinTransformer, SwinIR, Restormer, and HINet models where the first three models prioritize classification, whereas R-Vi-Swin-Net performs diagnostic labeling on visual samples via introduced models in various quantization levels. The proposed framework attained a confidence level exceeding 95% while manifesting competitive PSNR, SSIM, and the metric described in ${SSIM}_{G}$ [5], consolidating robust and reliable image classification in the clinical domain of malignant and benign lesions with attained a peak accuracy of 97.87 % at the 22nd epoch; furthermore, refined images afford profound fidelity in oral lesions with restored structural features, such as edges and textures, thereby catalyzing improved health monitoring. The reported results are obtained on a compact dataset of 750 images and should therefore be regarded as a proof-of-concept; further validation on larger, multi-centre cohorts is required before any clinical generalisation can be claimed.