A sparse-coding based approach to clothing image retrieval

In this paper, we present a sparse-coding based clothing image retrieval method. Our proposed method utilizes multiple types of low and high-level features such as clothing type, color, and appearance to describe an input clothing image. Based on the recent success of sparse representation, we advance a locality-sensitive sparse coding framework on the derived features for retrieving relevant instances from a clothing image collection. Compared with prior image retrieval or recommendation methods which either aimed at determining a proper similarity measure or required the knowledge or preference of prior users, our sparse-coding based approach is able to identify the most similar data instances based on its content information. From our experimental results on a real-world commercial clothing image dataset, we not only verify the effectiveness of our proposed framework, we also confirm that our approach outperforms baseline and state-of-the-art clothing image retrieval methods.