Visual search has become an invaluable tool for consumers looking to find similar products through images. This paper introduces a novel approach to enhance image search in the e-commerce domain by leveraging advanced deep learning techniques. Our method integrates U \(^{2}\) -Net to perform accurate product image segmentation, effectively delineating product boundaries and isolating them from the background. After segmentation, Bootstrapping Language-Image Pre-training (BLIP) is employed to generate detailed and descriptive captions that capture key product attributes, enhancing semantic understanding. This combination of segmentation and captioning not only improves the system’s ability to comprehend product details but also mitigates the impact of noise and background clutter, leading to a more accurate representation of the product. The proposed method offers significant improvements in search precision and retrieval relevance, making it a valuable addition to the e-commerce search ecosystem. We implemented the proposed models within a Gradio-based web framework to ensure a seamless and user-friendly experience for the final system implementation. This interface allows users to easily interact with the visual search system, making it accessible and intuitive while showcasing the enhanced capabilities of the deep learning models in a practical, real-world setting.