Performance Evaluation of Enhanced Deduplication Model with Image Augmentation using Deep Learning (IDME-IR)

Abstract

This study aims to evaluate the performance of the Image Duplicate Matching and Elimination - Image Retrieval (IDME-IR) Deduplication Model and an image augmentation method for image retrieval across key parameters such as accuracy, precision, recall, and F1-score. The IDME-IR Deduplication Model focuses on eliminating redundant or near-duplicate images from large datasets, ensuring a cleaner and more efficient retrieval process. Meanwhile, image augmentation techniques are employed to enhance dataset diversity, improving the robustness of retrieval systems by simulating real-world variations in lighting, orientation, and noise. Both methods are evaluated within the context of image retrieval tasks, with the performance metrics being computed across various datasets. After Evaluating the Performance of IDME-IR with Image Augmentation we get accuracy 93.55%, Precision 92.1%, F1-Score 92.7% and Recall 93%. Similarly, without applying image augmentation techniques the result has been observed as accuracy 84.2%, Precision 85.3%, F1-Score 84.7% and Recall 85.4%.