Published Paper Details:

The task of producing accurate and contextually relevant textual descriptions from visual inputs has attracted a lot of attention in the field of image captioning. This research proposal uses the Flickr8k dataset as a baseline for evaluation to tackle this difficulty in the context of Tamil language. In order to better understand the relative advantages and trade-offs of each approach, the study examines three different methods for improving Tamil image captioning. This helps to clarify the challenges associated with non-English captioning assignments. Using Long Short-Term Memory (LSTM) networks and the Convolutional Neural Network (CNN) VGG16, the first method creates a baseline. This methodology leverages the sequential data processing capability of LSTM and the visual feature extraction skill of VGG16 to provide a basis for further comparisons. The paper then presents an attention-based encoder-decoder architecture that adds an attention mechanism to LSTM layers. During caption synthesis, this method dynamically selects pertinent image segments to increase caption accuracy. Assessment metrics are employed to evaluate the model's effectiveness in producing contextually accurate captions, including BLEU scores and attention plots. Finally, by combining Vision Transformers (ViTs) with Transformer Networks for image processing and sequence production, respectively, the research pushes the envelope of innovation. Using Transformer Networks' and ViTs' complementing strengths, this sophisticated approach aims to dramatically improve caption quality by understanding the complex relationship between visual content and spoken descriptions. The research clarifies the trade-offs related to striking a balance between model complexity, computing demands, and caption authenticity through extensive comparison analysis. Moreover, it highlights the linguistic complexities present in Tamil, a language distinguished by its intricate syntactic and morphological structure. This research adds to the expanding body of information on multilingual picture captioning and offers critical insights and guidance for future developments in linguistically sensitive captioning systems that can function well in a variety of linguistic and cultural situations.