Adaptive Skin Color Model for Clothing Genre Recognition via Particle Swarm Optimization

Clothing genre recognition has shown its capabilities in many intelligent fashion scenarios. Given an unannotated consumer photo of a full-body person, the proposed study addresses the problem of recognizing the upperwear genres presented in that photo. Although the topic continues to show progress, most of the existing studies suffered from weaknesses related to skin identification. Therefore, to achieve this goal, we exploit the visual style elements of clothes to capture the discriminative attributes of each clothing genre and utilize an adaptive skin color model based on hill-climbing segmentation and Particle Swarm Optimization (PSO) to identify the skin color. The experimental results show that integrating these two approaches into a clothing recognition framework can lead to significant improvements over baselines, achieving new state-of-the-art results. Importantly, our method achieves these satisfactory results with a compact representation that does not require a large amount of training data to generate.