This paper proposes a novel image captioning algorithm based on a semantic attention model that combines top-down and bottom-up approaches. The model uses semantic attention to selectively focus on important visual concepts and fuse them into hidden states and outputs of recurrent neural networks (RNNs). This feedback mechanism connects top-down and bottom-up processing, enabling the model to generate more accurate and coherent image descriptions. The algorithm is evaluated on two public benchmarks: Microsoft COCO and Flickr30K. Experimental results show that the proposed method significantly outperforms state-of-the-art approaches across various evaluation metrics, including BLEU, Meteor, and CIDEr. The model's key contributions include the use of semantic attention to dynamically select and combine visual concepts, and the integration of both top-down and bottom-up information within an RNN framework. The model also incorporates a feedback loop that allows the RNN to refine its predictions based on attention weights from both input and output layers. The paper also discusses the role of visual attributes in caption generation and presents results from experiments comparing different attribute detection and attention schemes. The proposed method achieves state-of-the-art performance on both MS-COCO and Flickr30K datasets, demonstrating the effectiveness of the semantic attention approach in image captioning.This paper proposes a novel image captioning algorithm based on a semantic attention model that combines top-down and bottom-up approaches. The model uses semantic attention to selectively focus on important visual concepts and fuse them into hidden states and outputs of recurrent neural networks (RNNs). This feedback mechanism connects top-down and bottom-up processing, enabling the model to generate more accurate and coherent image descriptions. The algorithm is evaluated on two public benchmarks: Microsoft COCO and Flickr30K. Experimental results show that the proposed method significantly outperforms state-of-the-art approaches across various evaluation metrics, including BLEU, Meteor, and CIDEr. The model's key contributions include the use of semantic attention to dynamically select and combine visual concepts, and the integration of both top-down and bottom-up information within an RNN framework. The model also incorporates a feedback loop that allows the RNN to refine its predictions based on attention weights from both input and output layers. The paper also discusses the role of visual attributes in caption generation and presents results from experiments comparing different attribute detection and attention schemes. The proposed method achieves state-of-the-art performance on both MS-COCO and Flickr30K datasets, demonstrating the effectiveness of the semantic attention approach in image captioning.