The presence of illumination variation in dermatological images has a negative impact on the automatic detection and analysis of cutaneous lesions. This paper proposes a new illumination modeling and chromophore identication method to correct lighting variation in skin lesion images, as well as to extract melanin and hemoglobin concentrations of human skin, based on an adaptive bilateral decomposition and a weighted polynomial curve tting, with the knowledge of a multi-layered skin model. Different from state-of-the-art approaches based on the Lambert law, the proposed method, considering both specular reection and diffuse reection of the skin, enables us to address highlight and strong shading effects usually existing in skin color images captured in an uncontrolled environment. The derived melanin and hemoglobin indices, directly relating to the pathological tissue conditions, tend to be less inuenced by external imaging factors and are more efcient in describing pigmentation distributions. Experiments show that the proposed method gave better visual results and superior lesion segmentation, when compared to two other illumination correction algorithms, both designed specically for dermatological images. For computer-aided diagnosis of melanoma, sensitivity achieves 85.52% when using our chromophore descriptors, which is 8~20% higher than those derived from other color descriptors. This demonstrates the benet of the proposed method for automatic skin disease analysis.