Image gradient based 3D roughness estimation and rendering for haptic palpation from a single skin image
Tutor / director / evaluatorKim, Kwangtaek
Document typeMaster thesis
Rights accessOpen Access
Background/purpose: Skin palpation and property analysis (roughness, dryness, stiffness, temperature) are crucial for skin examination and diagnose. That is why is needed a noncontact-based method that allows to carry it out avoiding secondary infections or damage. A haptic device with haptic feedback was designed some years ago, but accurate results for 3D skin surface reconstruction and roughness estimation are still in research and improvement. In this study is proposed a gradient-based skin surface 3D roughness estimation algorithm that will enable haptic palpation and roughness examination. Methods: 3D roughness is estimated from 2D single image. First step is pre-processing the image, to improve the quality and reduce the noise by using contrast stretching and bilateral filtering. After, the gradient field is computed and used to obtain the 3D surface reconstruction using a surface-from-gradient algorithm, which will allow 3D roughness computation for a later dynamic haptic rendering. Results: Texture and curvature of the 3D reconstructed surface are checked in the first experiment, comparing roughness and geometry errors between a reconstructed surface using the proposed algorithm and two other algorithms, as well with a ground truth surface. The second experiment tests the method using in-vivo real skin disease images to compute roughness estimation and decomposition and also tasting haptic rendering in a haptic device. The experimental results verify the validity of our method. Conclusion: Roughness is a crucial property for dermatologists to examine skin disease (e.g., cases of psoriasis, atopic eczema or aging), that is why the proposed method of roughness estimation for haptic rendering will be extremely useful for dermatologists, improving the skin diagnose. In addition, the proposed method does not require complex medical systems to be implemented, since single image reconstruction is used.