Beyond hue labels: a data-driven analysis of lightness–chroma variation in online fashion assortments

Abstract

Online fashion platforms commonly organize products using broad hue labels such as red, blue, green, and yellow, although products within the same hue category may vary substantially in lightness and chroma. This study examines how within-hue color variation in online fashion assortments can be characterized using lightness–chroma (L*–C*) coordinates derived from the Commission Internationale de l’Éclairage L*a*b* color space (CIELAB). Product images were collected from a major South Korean online fashion retail platform using four hue-based knitwear queries, with 200 images retained for each hue query (N = 800). Garment regions were isolated using semantic segmentation, and dominant garment colors were extracted from CIELAB pixel values using MiniBatch K-means. The resulting L* and C* coordinates were analyzed using distributional analysis, data-adaptive K-means partitioning, and principal component analysis. Products retrieved under the same hue-based query occupied broad and differently configured regions of L*–C* space, with the four assortments differing in their lightness and chroma ranges and regions of product concentration. Data-adaptive partitioning summarized these within-hue distributions into distinct cluster configurations. An exploratory comparison using an additional 800 images collected approximately six months later further showed that changes in distributional location, dispersion, and overall L*–C* composition varied across hue-specific assortments. These findings demonstrate that nominal hue labels provide only a partial representation of color variation within online fashion assortments. The proposed framework provides a quantitative approach for characterizing and monitoring within-hue lightness–chroma composition, with potential applications in fashion assortment auditing, color planning, and more refined digital product organization.