Markov-Gibbs Random Field Approach for Modeling of Skin Surface Textures

• Main Author: Batool, Nazr e
• Format: Final Year Project
• Language: English
• Published: Universiti Teknologi Petronas 2008
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://utpedia.utp.edu.my/7560/1/2008-Markov-Gibbs%20Random%20Field%20Approach%20For%20Modeling%20Of%20Skin%20Surface%20Texture.pdf; http://utpedia.utp.edu.my/7560/

Medical imaging has been contributing to dermatology by providing computer-based assistance by 2D digital imaging of skin and processing of images. Skin imaging can be more effective by inclusion of 3D skin features. Furthermore, clinical examination of skin consists of both visual and tactile inspection. The tactile sensation is related to 3D surface profiles and mechanical parameters. The 3D imaging of skin can also be integrated with haptic technology for computer-based tactile inspection. The research objective of this work is to model 3D surface textures of skin. A 3D image acquisition set up capturing skin surface textures at high resolution (~0.1 mm) has been used. An algorithm to extract 2D grayscale texture (height map) from 3D texture has been presented. The extracted 2D textures are then modeled using Markov-Gibbs random field (MGRF) modeling technique. The modeling results for MGRF model depend on input texture characteristics. The homogeneous, spatially invariant texture patterns are modeled successfully. From the observation of skin samples, we classify three key features of3D skin profiles i.e. curvature of underlying limb, wrinkles/line like features and fine textures. The skin samples are distributed in three input sets to see the MGRF model's response to each of these 3D features. First set consists of all three features. Second set is obtained after elimination of curvature and contains both wrinkle/line like features and fine textures. Third set is also obtained after elimination of curvature but consists of fine textures only. MGRF modeling for set I did not result in any visual similarity. Hence the curvature of underlying limbs cannot be modeled successfully and makes an inhomogeneous feature. For set 2 the wrinkle/line like features can be modeled with low/medium visual similarity depending on the spatial invariance. The results for set 3 show that fine textures of skin are almost always modeled successfully with medium/high visual similarity and make a homogeneous feature. We conclude that the MGRF model is able to model fine textures of skin successfully which are on scale of~ 0.1 mm. The surface profiles on this resolution can provide haptic sensation of roughness and friction. Therefore fine textures can be an important clue to different skin conditions perceived through tactile inspection via a haptic device.