Enhancing Image Understanding: A Comparative Analysis of Encoder Architectures for Image Captioning

Image caption generation is a critical research area that combines computer vision and natural language processing, with wide-ranging implications such as assisting visually impaired individuals, improving autonomous vehicle capabilities, and refining image search algorithms. The main goal of image captioning is to identify objects in an image and explain the relationships between them using text. In the context of image captioning, this research compares three popular encoding architectures: ResNet50, VGG16 and InceptionV3. To analyse the input image in order to obtain significant features that are then converted for vector representation, an Encoder component of the architecture is used. These encoded features are used by the decoder module to create a coherent text description using the Long Short Term Memory LSTM model. By conducting thorough experiments and evaluations, this study aims to uncover the performance differences among the mentioned encoder architectures and determine the best model for image captioning tasks.