Objective To evaluate the availability of publicly available skin disease image datasets for image recognition research of hyperpigmentation disorders, and identify data sources that can be directly applied to supervised learning, external validation and interpretability analysis. Methods Public dataset platforms, official dataset websites and original publications were searched using keywords including dermatological image, hyperpigmentation, melasma, dermatology image dataset and skin tone. Datasets with publicly available or accessible upon license application that provide dermatological images, disease labels or lesion concept labels were included. Indicators such as dataset image scale, label structure, hyperpigmentation-related labels, skin tone/skin type metadata, annotation sources, licensing modes and applicability for image recognition research were extracted. Descriptive statistics were used to analyze dataset composition, and class imbalance analysis was performed on the Mendeley SDC Dataset with clear category quantities. Results A total of four publicly available datasets were included: SCIN, Mendeley SDC Dataset, Fitzpatrick17k and SKINCON. The Mendeley SDC dataset contained 9548 images, among which 700 were hyperpigmentation-related images,accounting for 7.33%. The ratio of the maximum sample size to the minimum sample size among 5 categories was 4.52. The SCIN dataset contained 10,408 voluntarily contributed dermatological images with dermatologist-assigned labels, symptoms and skin tone/skin type information. The Fitzpatrick17k dataset contained 16,577 clinical dermatological images and labels of 114 disease categories, and provided Fitzpatrick skin type annotations. SKINCON provided 48 clinical concept annotations of skin lesions based on subsets of Fitzpatrick17k and DDI. Conclusion Among existing public datasets, the Mendeley SDC dataset is most suitable as initial supervised training data for hyperpigmentation recognition tasks. SCIN and Fitzpatrick17k are more appropriate for external validation and skin tone-stratified evaluation. SKINCON is applicable for model interpretability and error analysis. Public datasets can offer a reproducible data foundation for image recognition research of hyperpigmentation disorders. Secondary verification of disease labels and external validation on local populations are still required before their application in clinical auxiliary diagnosis of melasma.