Developing an advanced machine learning model for identifying virtual montages in digital media

Abstract

Virtual montages are compositions that blend several digital materials to produce a new, visually appealing narrative and are becoming increasingly popular, as a result of the quick development of digital media. Virtual montages have emerged as a potent technique for improving visual storytelling and content presentation across platforms in the digital age. However, the quick expansion of digital content has made advanced methods for effectively recognizing and classifying these montages necessary. This work suggests a novel method for virtual montage identification across a broad range of visual themes, content types, and media resolutions, and resolutions utilizing an Adaptive Flower Pollination Optimized Mutual Information with Naïve Bayes (AFPO-MI-NB) model. The research’s dataset comprises a variety of content from virtual montage detection dataset which enables the model to generalize effectively to data from the real world. To improve the model’s capacity to handle various image and video qualities, pre-processing methods, including data augmentation and pixel value normalization are used. Convolutional neural networks (CNN) are used for feature extraction to capture spatial patterns. The model uses AFPO-MI-NB classification to enhance classification accuracy and computational efficiency. AFPO allows for more adaptable feature weights, MI evaluates the relationship between visual features and the classification task, and NB classifier processes feature matrices for binary classification. The hybrid approach strengthens feature selection through global and local searches, resulting in a model that enhances classification accuracy and improves computational efficiency. According to the experimental data, this model provides a reliable solution for virtual montage identification across a variety of media formats, outperforming current techniques in terms of F1-score of 0.95, recall of 0.92, and precision of 0.97. This work has significant applications in automated digital media analysis, copyright enforcement, and content control.