Scholarlink

<p>In this work, a hybrid deep learning framework is proposed for image captioning through the combination of Attention-based CNN (convolutional neural network) for feature extraction and LSTM (long short-term memory) for sequential data generation. This approach provides a mechanism to connect the understanding of the visual world with that of natural language, allowing automatic language generation of captions for images. By employing hybrid approach, CNNs extract visual information from the images, while the LSTM units then interpret these features to produce coherent and contextually appropriate textual outputs. Trained on MS COCO dataset, the proposed framework attains the highest scores based on evaluation metrics to assess the quality of machine generated text. The framework attains BLEU-4 (bilingual evaluation understudy) score of 35.8, METEOR (metric for evaluation of translation with explicit ordering) score of 29.5, ROUGE-L (recall oriented understudy for gisting evaluation) score of 58.6 and CIDEr (consensus-based image description evaluation) score of 97.2 thereby surpassing existing baseline models. The results from qualitative analysis indicate that generated captions not only matched human-level interpretations but also capture nuanced contextual relationships between objects.</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 hybrid deep learning framework to generate language for image captioning

  • 期刊论文

  • 发表日期：

    2025年8月29日

查看全文
  • Pawan Bhambu,
  • Sunil Kumar,
  • Kavita Arora,
  • Saurav Chandra,
  • Vibhu Sharma,
  • Dattatreya Mankame,
  • Bhim Sain Singla

摘要

In this work, a hybrid deep learning framework is proposed for image captioning through the combination of Attention-based CNN (convolutional neural network) for feature extraction and LSTM (long short-term memory) for sequential data generation. This approach provides a mechanism to connect the understanding of the visual world with that of natural language, allowing automatic language generation of captions for images. By employing hybrid approach, CNNs extract visual information from the images, while the LSTM units then interpret these features to produce coherent and contextually appropriate textual outputs. Trained on MS COCO dataset, the proposed framework attains the highest scores based on evaluation metrics to assess the quality of machine generated text. The framework attains BLEU-4 (bilingual evaluation understudy) score of 35.8, METEOR (metric for evaluation of translation with explicit ordering) score of 29.5, ROUGE-L (recall oriented understudy for gisting evaluation) score of 58.6 and CIDEr (consensus-based image description evaluation) score of 97.2 thereby surpassing existing baseline models. The results from qualitative analysis indicate that generated captions not only matched human-level interpretations but also capture nuanced contextual relationships between objects.