Year: 2025
Type: conference-paper
Abstract: This paper presents a deep learning approach forgenerating descriptive captions for images through the combination of deep learning driven image understanding and textgeneration. The rapid progress in deep learning has greatlyenhanced machines’ capacity to interpret visual information andgenerate natural-language descriptions. Among these advancements, Transformer architectures have proven especially effectivedue to their self-attention mechanisms, which enable modelsto better capture global image, text relationships. Traditionalimage captioning models that use CNNs for visual encoding andRNNs for sentence generation often struggle to model long-rangedependencies and produce contextually rich descriptions.To address these challenges, this study introduces an intelligent image captioning framework that utilizes InceptionV3 forvisual feature extraction and a custom-built Transformer-basedencoder decoder architecture for generating textual descriptions.Unlike conventional systems that depend on pre-trained languagemodels, the decoder in this work is implemented and optimizedexclusively for image captioning. Comprehensive training andevaluation are conducted using a large-scale benchmark dataset,MS COCO, to validate the system’s effectiveness across diverseimage domains. This paper aims to integrate image understanding and linguistic reasoning by providing a practical frameworkwith applications in accessibility, digital asset management, andautomated content generation.
Institution: Nepal College of Information Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7118176499