This paper presents a guided visual attention (GVA) approach for automatic image caption generation. The GVA approach incorporates an additional attention mechanism to re-adjust the attentional weights on the visual feature vectors and feed the resulting context vector to the language LSTM. The first-level attention module is used as guidance for the GVA module, and re-weighting the attention weights significantly enhances the caption's quality. The encoder-decoder architecture is used, where Faster R-CNN is used for extracting features in the encoder and a visual attention-based LSTM is applied in the decoder. Extensive experiments were conducted on the MS-COCO and Flickr30k datasets. The proposed approach achieved an average improvement of 2.4% on BLEU@1 and 13.24% on CIDEr for the MS-COCO dataset, as well as 4.6% on BLEU@1 and 12.48% on CIDEr for the Flickr30K datasets. These results demonstrate the clear superiority of the proposed approach compared to existing methods. The implementing code can be found at https://github.com/mdbipu/GVA. Image captioning is a challenging task that involves transforming a series of pixels into a list of words that convey meaningful descriptions of the image content. This task can be viewed as a multidimensional image classification problem that aims to automatically assign captions to an image by leveraging the relationships between the extracted visual features and the caption words. Image captioning has many beneficial applications, including human-computer interaction, visual question answering, biomedical imaging, robotics, and assistive technologies for people who can't see. Several promising techniques for image captioning have been introduced, including template-based techniques, neural-based approaches, attention-based strategies, reinforcement learning-based frameworks, and transformer-based methods. The attention-based approach is one of the most rapidly progressing ways to make image captioning tasks more convenient. These days, most people focus on exploiting the encoder-decoder framework that typically uses a CNN to encode the image features and an RNN to decode the sentences. In addition to this framework, the attention mechanism has become a basic tool for model learning, and more scholars are using this mechanism to improve their model performance. Visual attention in image captioning refers to the ability of a deep neural network model to identify and emphasize relevant aspects, resulting in a more accurate and informative description of the visual content. This is particularly important when the image contains multiple visual objects or regions. By using visual attention, image captioning models can generate more accurate and detailed captions, as they can focus on an image's essential regions and ignore extraneous regions. However, there are still several limitations and challenges associated with this approach, such as attention drift, where the model focuses on irrelevant or incorrect regions of the image, leading to inaccurate captions.This paper presents a guided visual attention (GVA) approach for automatic image caption generation. The GVA approach incorporates an additional attention mechanism to re-adjust the attentional weights on the visual feature vectors and feed the resulting context vector to the language LSTM. The first-level attention module is used as guidance for the GVA module, and re-weighting the attention weights significantly enhances the caption's quality. The encoder-decoder architecture is used, where Faster R-CNN is used for extracting features in the encoder and a visual attention-based LSTM is applied in the decoder. Extensive experiments were conducted on the MS-COCO and Flickr30k datasets. The proposed approach achieved an average improvement of 2.4% on BLEU@1 and 13.24% on CIDEr for the MS-COCO dataset, as well as 4.6% on BLEU@1 and 12.48% on CIDEr for the Flickr30K datasets. These results demonstrate the clear superiority of the proposed approach compared to existing methods. The implementing code can be found at https://github.com/mdbipu/GVA. Image captioning is a challenging task that involves transforming a series of pixels into a list of words that convey meaningful descriptions of the image content. This task can be viewed as a multidimensional image classification problem that aims to automatically assign captions to an image by leveraging the relationships between the extracted visual features and the caption words. Image captioning has many beneficial applications, including human-computer interaction, visual question answering, biomedical imaging, robotics, and assistive technologies for people who can't see. Several promising techniques for image captioning have been introduced, including template-based techniques, neural-based approaches, attention-based strategies, reinforcement learning-based frameworks, and transformer-based methods. The attention-based approach is one of the most rapidly progressing ways to make image captioning tasks more convenient. These days, most people focus on exploiting the encoder-decoder framework that typically uses a CNN to encode the image features and an RNN to decode the sentences. In addition to this framework, the attention mechanism has become a basic tool for model learning, and more scholars are using this mechanism to improve their model performance. Visual attention in image captioning refers to the ability of a deep neural network model to identify and emphasize relevant aspects, resulting in a more accurate and informative description of the visual content. This is particularly important when the image contains multiple visual objects or regions. By using visual attention, image captioning models can generate more accurate and detailed captions, as they can focus on an image's essential regions and ignore extraneous regions. However, there are still several limitations and challenges associated with this approach, such as attention drift, where the model focuses on irrelevant or incorrect regions of the image, leading to inaccurate captions.