Deep Neural Networks for Image Caption Generation
K. Tejaswi, Mohan Lal · International Journal of Research Publication and Reviews · 2025
This project explores the rapidly growing field of image caption generation, which merges the strengths of computer vision and natural language processing to automatically produce human-like textual descriptions of images.The proposed framework utilizes Convolutional Neural Networks (CNNs) for extracting deep visual features and Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM) units for generating fluent and contextually meaningful captions.Trained on large-scale benchmark datasets, the model effectively captures complex visual-semantic relationships, enabling accurate and grammatically coherent caption generation.Building upon this core model, an Advanced Image Caption Generator web application is developed, designed to deliver 4-5 optimized captions per image, while also offering advanced capabilities such as object, color, and scene detection.To increase accessibility and usability, the system supports real-time translation into more than 40 languages, with a particular focus on Indian languages such as Telugu and Hindi.Optimized for performance, the application processes each image in just 3-4 seconds, combining speed, accuracy, and a user-friendly interface to provide a seamless experience.Experimental results demonstrate that this approach achieves competitive performance when compared with existing state-of-the-art models, showing significant improvements in accuracy, fluency, and multilingual adaptability.This work highlights the potential of deep neural networks in real-world applications, including digital accessibility for the visually impaired, intelligent image retrieval, content management, and assistive technologies, while paving the way for future advancements in multimodal artificial intelligence systems.