Image Caption Generation using ResNET-50 and LSTM
Satish Kumar Satti, Goluguri N. V. Rajareddy, Prasad Maddula, N V Vishnumurthy Ravipati
Abstract
Generating image captions presents a formidable challenge, requiring a deep understanding of image content to produce coherent and descriptive textual descriptions. This research introduces an innovative approach for image caption generation, leveraging the ResNet-50 convolutional neural network (CNN) and the Long Short-Term Memory (LSTM) recurrent neural network. The ResNet-50 CNN serves as a robust feature extractor, encoding essential visual information from the input image. Through fine-tuning on a comprehensive dataset, the pre-trained ResNet-50 model becomes proficient at capturing the semantic essence of images. We employ an LSTM network as a language model to craft textual descriptions. It inputs the encoded ResNet-50 features and progressively generates captions word by word. Our LSTM model is trained on an extensive collection of image-caption pairs, enabling it to grasp the intricate connections between visual attributes and corresponding text descriptions. Empirical results unequivocally demonstrate that our proposed model surpasses existing methods in terms of caption quality and semantic coherence.