Semantic-Aware Image Deduplication: Leveraging Object Recognition for Enhanced Accuracy

Image deduplication is an important process when working with large scale image sets, which finds its application in various areas, starting with censorship and extending to storage optimization.Conventional approaches often rely on pixel-based feature matching or perceptual hashing techniques, which fail to capture the semantic similarity of images.This paper introduces a new approach to image deduplication to be based on the object recognition mechanism to provide information to be used in the process of deduplication.By leveraging deep learning techniques for object detection and classification, the proposed method allows for an increased level of precision in detecting similar and identical images even in case of considerable variations of the visual features.The performance evaluation of the proposed semantic-aware deduplication framework demonstrates an F1-score of 0.947 and a mean Average Precision (mAP) of 0.939 on the MNIST-Duplicate dataset.The framework achieves improvements ranging from 2% to 4% over optimized deep learning baselines across several benchmark datasets.Incorporating semantic understanding into image deduplication enhances accuracy and overall effectiveness, paving the way for broader adoption of intelligent systems in the field of image management.