A Comparative Study of Feature Extraction Models for Image Caption Generation
Yugant Gotmare · International Journal for Research in Applied Science and Engineering Technology · 2024
Abstract: Image caption generation is a challenging task in the field of computer vision and natural language processing. This study presents a comparative analysis of various feature extraction models for image caption generation. The goal is to evaluate the performance and effectiveness of different models in capturing visual features and generating accurate and contextually relevant captions. The feature extraction models considered in this study include ResNet (Residual Neural Network), DenseNet, VGG (Visual Geometry Group), InceptionNet, DenseNet, and XceptionNet. We conduct extensive experiments on the Flickr 8k dataset, utilizing a common architecture for the caption generation task. Evaluation is performed using BLEU scores to assess the quality and similarity of generated captions to human references. Our results indicate varying performance across the different models, with notable differences in BLEU scores. We further discuss the strengths and limitations of each model, providing insights into their suitability for image caption generation. This comparative study contributes to the understanding of the impact of feature extraction models on the performance of image caption generation systems.