Enhanced Shot Boundary Detection Through Hybrid Video Segmentation Techniques

Detecting shot boundaries, a crucial aspect of video analysis, allows for effective video organization and search through recognizing changes between consecutive shots. In this work, a new method that integrates Wavelet Transform, EfficientNet, and Deep Neural Networks (DNN) to improve the precision of SBD, especially for identifying small changes. This approach is tested on the TRECVid 2017 dataset and shows a significant enhancement compared to current models, reaching 80% accuracy rate overall which shows a 15% enhancement when compared to standard DNN models. Additionally, the performance metrics such as precision, recall and F1 Score, are calculated for three categories: No-transition, abrupt transition and subtle transition. The combined model obtained a precision, recall and F1 score of Abrupt Transition as 0.72, 0.74, 0.70, and for Subtle transition, which is usually complicate to detect, has achieved a efficiency of 0.80, 0.80, 0.78 for precision, recall and F1 score respectively. The model has also performed well with no-transitions, providing the results as 0.79, 0.78, and 0.73 for precision, recall, F1 score respectively. This study shows that the proposed method is successful in identifying sudden and gradual changes, improving the accuracy of video shot boundary detection