The rapid expansion of multimedia repositories surpasses the capabilities of traditional data analysis. An advanced image retrieval (IR) model is essential for accurately retrieving relevant images. Numerous machine learning and deep learning models have been proposed to address this challenge effectively. Even though the deep learning models are so powerful in extracting the image features, they are getting failed in prioritizing the most informative parts of the feature maps and also ignored the focus on specific channels or spatial regions. This will lead to the less effective feature representations of images, which contain varying levels of importance across different regions or channels. As a result, the retrieval performance may suffer because the model may not effectively differentiate between similar images. To address this point, this paper concentrated on refining the feature maps by applying channel and spatial attention, which helps in highlighting the most informative features and suppressing less important ones. This has become possible by integrating the Convolutional block of attention module (CBAM) and ResNet-50 models. CBAM is integrated with ResNet-50 by adding attention modules after each residual block. These modules apply channel and spatial attention mechanisms to the feature maps, refining them before passing to the next layer. The proposed model is verified on benchmark databases Corel-10k, CIFAR. The significant performance of the proposed method shown over existing deep learning models.