Modelling of Firefly Algorithm with Densely Connected Networks for Near-Duplicate Image Detection System

Near-duplicate image detection is the way of detecting and flagging images that are highly similar to each other but not identical. It is a crucial task in different fields, including search engines, content management, and copyright enforcement, as it helps in content and organization de duplication. To achieve this, techniques and algorithms are employed that calculate similarity metrics, compare features, or analyze image content to define the degree of resemblance between images. Near-duplicate image detection can include approaches such as feature extraction, machine learning, and perceptual hashing to effectively manage and identify similar images in large datasets, which offer benefits in content retrieval and storage optimization. Therefore, this study presents a new firefly algorithm with deep learning-based near-duplicate image detection (FFADL-NDID) technique. Initially, median filtering (MF) approach is used to preprocess the input images. The proposed FFADL-NDID technique exploits the robust feature extraction abilities of DenseNet, a pre-trained DL model for capturing complex visual patterns from database and query images. In addition, the FFA is applied to carry out the hyperparameter tuning, optimizing the system for superior performance. This synergistic fusion enhances the overall efficiency of near-duplicate image detection by successfully searching the hyperparameter space for optimal configurations. Finally, the FFADL-NDID framework applies Euclidean distance-based similarity matching processes, which detects the near-duplicate images significantly. The simulation analysis of the FFADL-NDID method is tested on multiple datasets and the outcomes show its promising performance over other DL models in terms of different measures.