Image caption generation is a crucial task at the intersection of computer vision and natural language processing (NLP), enabling automated generation of meaningful descriptions for images. While existing approaches leverage deep learning, many models struggle with generating contextually rich and accurate captions. This paper presents an optimized deep learning-based image captioning model that integrates DenseNet for visual feature extraction with Long Short-Term Memory (LSTM) networks for sequential caption generation. DenseNet’s dense connectivity structure allows for enhanced feature reuse, improving the extraction of intricate visual details. These extracted features are then processed by an LSTM-based decoder, which generates captions by predicting words sequentially in a sequence-to-sequence learning framework. To improve training efficiency and scalability, we implement a custom data generator that efficiently processes large-scale datasets. The proposed model is trained and evaluated on benchmark datasets, using standard performance metrics such as BLEU, and CIDEr to assess caption accuracy and fluency. Experimental results demonstrate that the combination of DenseNet and LSTM significantly enhances caption quality, outperforming traditional CNN-RNN architectures. This study contributes to advancements in automated content generation, image retrieval, and AI-driven accessibility solutions.