Altered Image Detection in E-commerce Products Using Feature Fusion Model

In today’s world, e-commerce websites have become the go-to choice for shoppers. Unfortunately, scammers exploit this by using altered or filtered photos to deceive customers, leading to dissatisfaction and financial losses. This study addresses this issue by introducing a novel approach for detecting altered images on e-commerce platforms. The proposed model employs a feature-level fusion method that combines traditional feature extraction with a Convolutional Neural Network (CNN)-based model to identify manipulated images effectively. This approach leverages the strengths of both traditional image processing techniques and modern deep learning architectures. Extensive experimentation has shown that this model significantly enhances accuracy, achieving a $\mathbf{9 5 \%}$ improvement over standalone CNN models. The model’s robustness and accuracy have been validated across various product categories, confirming its suitability for different types of e-commerce images. The primary objective is to increase the reliability of product images, thereby boosting customer satisfaction and reducing return rates.