The evolution of e-commerce platforms in recent years has led to increased use of recommendation systems while maintaining the customer-product relationship. Artificial intelligence (AI) plays a key role in creating intelligent platforms. Companies like Amazon, Myntra, and Adidas exemplify the effectiveness of these systems. Recommendation systems aim to suggest products based on customer experience and preferences. Visual recommendations focus on using multimedia content, such as images and videos, to enhance these suggestions. Image embedding enables the extraction of significant object features through deep neural networks. Similarity metrics, such as Cosine, Euclidean, Manhattan, and Jaccard distances, are essential for providing relevant results. In our study, we aim to explore the potential of these similarity metrics within a recommendation system. Utilizing the pre-trained VGG16 architecture, we compare these metrics over the same context, the experiment conducted on the test phase yielded impressive results, with performance scores of 86.80% in Avg_Acc@5, 85.57% in Avg_Acc@7, and 83.40% in Avg_Acc@10, resulting in a global average of 85.25% on the Fashion Image Product Dataset [1]. This research highlights the importance of similarity metrics in improving the relevance of visual recommendations in e-commerce.