Retrieving images from databases based on user queries has become more computationally expensive due to the growth of multimedia content and image processing techniques. Annotation-based image retrieval systems are ineffective because normal pixel-by-pixel matching of images leads to large changes in terms of storage, angle, and patterns. Content-Based Image Retrieval (CBIR) techniques have been developed to tackle this problem. By extracting and comparing relevant features, CBIR techniques properly quantify matches between query photos and database images. This manuscript proposes a novel CBIR system that combines deep learning and machine learning methods for retrieving color logo images. The system uses KNN for image comparison and pre-trained deep learning models VGG16 and ResNet50 for feature extraction. The proposed method achieves more precision than current CBIR systems by utilizing KNN and the Euclidean distance metric to evaluate the similarity between query photos and database images. This creative method is adaptable and can be used for a number of CBIR applications, including crime prevention, digital libraries, fingerprint identification, and historical research.