Content-Based Fashion Image Retrieval in Android Applications using Artificial Neural Networks

The rapid rise of e-commerce and digital fashion has increased demand for intelligent search systems. Traditional keyword-based approaches fail to accurately capture user preferences, often leading to poor recommendations. This research addresses this challenge by proposing a Content-Based Image Retrieval (CBIR) system tailored for fashion, deployed in a React-Native mobile application. The system uses deep learning models, including Convolutional Neural Networks (CNNs) and Artificial Neural Networks (ANNs), to extract detailed features from fashion images. These features are indexed and compared using similarity metrics like cosine similarity and Euclidean distance to retrieve relevant results. Autoencoders and attention mechanisms refine feature extraction, while contrastive and triplet loss functions improve embedding quality. Users can search by uploading fashion images, leading to a more intuitive shopping experience. The solution demonstrates high effectiveness, achieving a stable training and validation accuracy of approximately 95.99% across multiple epochs. This study proves the viability of deep learning-powered CBIR for fashion, improving mobile-based recommendations and offering reliable real-time results. The integration of efficient retrieval techniques, optimized indexing, and ANN-based models ensures the system is scalable and user-friendly for practical applications in fashion search.