Detecting Duplicate Products in E-Commerce Images Using Siamese Networks
Enis Teper, Furkan Eseoğlu, Mustafa Keskin · 2024
Although merchants typically sell a single product, they may inadvertently or deliberately include more than one of the same product in an image. Consequently, users might be uncertain whether they are purchasing a single item or multiple identical items. This study aims to eliminate such confusion. Initially, the system developed for this purpose will detect objects within the image. These detected objects are then converted into embeddings using a Siamese network. Once the objects have been converted into representation layers, the distances between them will be calculated. If the distance between any two objects is less than a predefined threshold, the image is identified as containing duplicate objects. By identifying and addressing images with duplicate products, this approach aims to reduce shopper confusion.