Abstract:Background: Skin tone and pigmented regions, associated with melanin and hemoglobin, are critical indicators of skin condition. While most prior research focuses on pigment analysis, the capability to simulate diverse pigmentation conditions could greatly broaden the range of applications. However, current methodologies have limitations in terms of numerical control and versatility. Methods: We introduce a hybrid technique that integrates optical methods with deep learning to produce skin tone and pigmented region-modified images with numerical control. The pigment discrimination model produces melanin, hemoglobin, and shading maps from skin images. The outputs are reconstructed into skin images using a forward problem-solving approach, with model training aimed at minimizing the discrepancy between the reconstructed and input images. By adjusting the melanin and hemoglobin maps, we create pigment-modified images, allowing precise control over changes in melanin and hemoglobin levels. Changes in pigmentation are quantified using the individual typology angle (ITA) for skin tone and melanin and erythema indices for pigmented regions, validating the intended modifications. Results: The pigment discrimination model achieved correlation coefficients with clinical equipment of 0.915 for melanin and 0.931 for hemoglobin. The alterations in the melanin and hemoglobin maps exhibit a proportional correlation with the ITA and pigment indices in both quantitative and qualitative assessments. Additionally, regions overlaying melanin and hemoglobin are demonstrated to verify independent adjustments. Conclusion: The proposed method offers an approach to generate modified images of skin tone and pigmented regions. Potential applications include visualizing alterations for clinical assessments, simulating the effects of skincare products, and generating datasets for deep learning.

Skin Image Synthesis Based on Physical Model

Estimation of Skin Pigment Concentration by Simulating Skin Color

Generation of skin tone and pigmented region-modified images using a pigment discrimination model trained with an optical approach

S-SYNTH: Knowledge-Based, Synthetic Generation of Skin Images

Gpu-Based Skin Texture Synthesis for Digital Human Model

Blind Separation of Skin Chromophores from Multispectral Dermatological Images

Deep learning-based pigment analysis model trained with optical approach and ground truth assistance

A Hyperspectral Space of Skin Tones for Inverse Rendering of Biophysical Skin Properties

Skin Color Modeling of Digital Photographic Images

Models to Predict Naturalness of Images Including Skin Colours

Spectrum-based deep learning framework for dermatological pigment analysis and simulation

Physical Measurement and Spectral Reproduction of Human Skin Color

Skin Tone Analysis Through Skin Tone Map Generation With Optical Approach and Deep Learning

Integrated deep learning approach for generating cross-polarized images and analyzing skin melanin and hemoglobin distributions

Analysis-by-synthesis Approach for Facial Modeling Based on Orthogonal Images

Incremental Elliptical Boundary Model for Accurate Skin Region Segmentation

A new approach for identifying skin diseases from dermatological RGB images using source separation

Skin Image Segmentation Based on Energy Transformation

Extracting intrinsic images from multi-spectral

Human Skin Detection In Images By Mser Analysis

New closed-form approximation for skin chromophore mapping