This project develops an advanced deep learning system for automated image captioning, focusing on unsupervised and semi-supervised techniques to minimize the reliance on large labeled datasets. Traditional methods for image captioning often require extensive human-labeled data to train models, which can be a significant barrier, especially in domains with limited annotated resources. This idea helps to easily identify the images that are difficult to address like medical images, satellite images, and many other fields where unlabeled data is found. Firstly, feature extraction is done by CNN, where the CNN collects and extracts the features from the images, and these features are used to train the algorithm. Then the GAN is used to generate and discriminate the captions accurately. Finally, scores like blue and meteor are used to measure the accuracy and quality of captions by this we can identify the algorithm’s performance and how efficiently the algorithm generates the captions. This project picks up the knowledge from unsupervised and semi-supervised learning where the future is running behind the complex unlabeled data and areas of unsupervised learning.