Using information features, content-based image retrieval (CBIR) retrieves and searches images from a vast database. In the past, various designs for thumb feature descriptors have been studied using visual cues like color, texture, and size to describe these images. Deep learning methods, however, been widely used as a substitute for the dominating design process for more than ten years. Using the data, characteristics are automatically learned. The interconnected dual deep convolutional neural network (IDD-CNN), which this study suggests, consists of two unique First CNN uses the characteristics to its advantage, and then custom CNN is made to take advantage of the unique qualities. Additionally, a unique directional network is created that consists of two blocks—a learning block and a memory block—that aid in determining picture resemblance. Because this study takes into account a sizable dataset, an ideal approach for compact features is presented. Additionally, IDD-CNN is assessed taking into account the two unique benchmark datasets. IDD-CNN performs better than the other current model on the Oxford dataset when mean average precision (mAP) measures and comparisons are taken into account.