Scholarlink

<p>Video summarization is a complex deep-learning application widely used in various video streaming platforms. As the available video content is enormous, video summarization helps analyze and classify it based on its content. Most applications try to automate the video summarization process, which helps search and analyze the videos in their library. Various research works have considered Bi-LSTM for the video summarization process, which uses more computational power than other dense networks. Hence, this paper proposes a supervised video summarization method using a deep learning model. This model summarizes the videos through key shots by utilizing a self-attention mechanism, which helps efficiently process videos and summarization. The self-attention mechanism uses the Bi-LSTM model for extracting the features. The LSTM model is of two types, one for video summarization (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({LSTM}_{VS}\)</EquationSource> </InlineEquation>) and another for finding HPP (Harmful Point Positioning) (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\({LSTM}_{HPP}\)</EquationSource> </InlineEquation>). The self-attention module comprises feed-forward and feed-backward passes for training. The proposed model is evaluated through TvSum and Summer datasets and other benchmark datasets to evaluate the model. The model with HPP provides a better overall prediction of 55.1, 61.8, and 58.7&#xa0;F-1 scores in the canonical, augmented, and transfer learning processes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Self-Attention and Memory Module for Video Summarization Using a Bi-long-short-term-memory Deep Learning Model

  • 期刊论文

  • 发表日期：

    2026年3月24日

查看全文
  • Asha Prashant Sathe,
  • P. Jeyanthi

摘要

Video summarization is a complex deep-learning application widely used in various video streaming platforms. As the available video content is enormous, video summarization helps analyze and classify it based on its content. Most applications try to automate the video summarization process, which helps search and analyze the videos in their library. Various research works have considered Bi-LSTM for the video summarization process, which uses more computational power than other dense networks. Hence, this paper proposes a supervised video summarization method using a deep learning model. This model summarizes the videos through key shots by utilizing a self-attention mechanism, which helps efficiently process videos and summarization. The self-attention mechanism uses the Bi-LSTM model for extracting the features. The LSTM model is of two types, one for video summarization ( \({LSTM}_{VS}\) ) and another for finding HPP (Harmful Point Positioning) ( \({LSTM}_{HPP}\) ). The self-attention module comprises feed-forward and feed-backward passes for training. The proposed model is evaluated through TvSum and Summer datasets and other benchmark datasets to evaluate the model. The model with HPP provides a better overall prediction of 55.1, 61.8, and 58.7 F-1 scores in the canonical, augmented, and transfer learning processes.