A Deep Learning And Meta-Heuristic Optimization Approach For Content-Based Video Retrieval

Abstract

Deep learning-based feature extraction has become a popular approach for Content-Based Video Retrieval (CBVR) due to its ability to capture the complex and discriminative features of the video data.Existing works in CBVR using deep learning-based feature extraction have focused on developing various deep learning models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), etc. for feature extraction from video frames. These models have been trained on large datasets to learn the underlying patterns and features in the video data.Deep learning models can be computationally expensive, especially for large and complex video datasets. This can limit the scalability and real-time performance of CBVR systems. Therefore, in this reserch work a novel CBVR model is intrdoced with the assistance acquired from the deep learning and meta-heuristic optimization model. The proposed model includes five major phases: Video pre-processing, Feature extraction, Feature selection, Video representation, Similarity computation.  In the Video pre-processing phase, the collected raw video is pre-processed via video-to-frame conversion, color correction via histogram equalization and denoising via median filtering. Then, visual and temporal features are extracted from the video frames using the new AlexNet and Recurrent Neural Network (RNN), respectively. Principal components of features are selected optimally via the new Self-Improved Snow Leopard Optimization Algorithm (SI-SLO). The proposed SI-SLO model is an extended version of the standard Snow Leopard Optimization (SLO). In the Video representation stage, the selected features are used to represent the video. Then, in the Similarity computation phase, the similarity between two videos is computed using the video representations.The videos can be ranked based on their similarity scores and displayed to the user. Theproposed model has been implemented in MATLAB. The evaluation has been made in terms of accuracy, precision, sensitivity, specificity as well.