Benchmarking Hybrid CNN-BiRNN Video Summarization Against Traditional Methods for Optimal Streaming
Asha Prashant Sathe, P. Jeyanthi · 2025
Video summarization improves the quality of the videos to be streamed online efficiently. Efficient video streaming should reduce the streaming time and cost with uncompromised video quality. Education, entertainment, surveillance systems, and other real-time applications have recently used video summarization to reduce streaming time. Several traditional video summarization methods cannot deliver complete video content and do not reduce the streaming time. This paper implements a Convolution Neural Network model incorporated with a Bi-RNN model to overcome these issues and develop an efficient video summarization process. CNN learns all the frames and extracts the visual and hidden information. The Bi-RNN compares the features recurrently with one another. Also, CNN acts as an encoder, and Bi-RNN acts as a decoder for comparing the frames to obtain contextually similar. The similarity is calculated using the Similarity Score Value function based on its score, and the frames selected from the video need to be summarized. The CNN and Bi-RNN are implemented in Python and experimented with the benchmark dataset SumMe, and TVSum to verify performance. The F1-score is calculated and compared with the other learning models for performance evaluation. From the comparison, it is noticed that the proposed model outperforms other models.