Accurate depth estimation of skin surface using a light-field camera toward dynamic haptic palpation

Area of Science:

• Biomedical Engineering
• Computer Vision
• Medical Imaging

Background:

• Haptic skin palpation requires 3D skin surface reconstruction for tactile and visual data acquisition.
• Accurate depth estimation from light field cameras is challenging due to low-resolution images and disparity matching errors.

Purpose of the Study:

• To develop an accurate 3D skin surface reconstruction method for haptic palpation.
• To overcome limitations in depth estimation from low-resolution light field camera images.

Main Methods:

• Employed Generative Adversarial Networks (GANs) for super-resolution of skin images, preserving surface details.
• Implemented a multi-step process for accurate skin depth estimation: lens distortion correction, sub-pixel image generation, cost-volume building, multi-label optimization, and refinement.

Main Results:

• The deep learning super-resolution method effectively preserved textural details like wrinkles, outperforming other methods.
• Computed depth maps demonstrated superior accuracy and robustness compared to existing state-of-the-art depth map computation techniques.

Conclusions:

• This study presents the first depth map estimation for skin surfaces using light field cameras.
• The proposed method shows significant superiority for 3D skin surface reconstruction and depth estimation.