A Method On Simple-user-Interaction-based Material Appearance Estimation From A Single Image

With the rapid developments of computer graphics and computer vision,advanced image editing technology is developing rapidly in recent years.More and more studies begin to pay close attention to the understanding of image content,such as the material and surface structure information of the model.We present an interactive system for modeling materials and normal from a single image.Given a texture image of a nearly planar surface under the natural light,our system models the detailed spatially-varying reflectance properties(diffuse,specular and roughness)and surface normal variations of each pixel with minimal user interaction,Than we can render the material model under any other light.The algorithm is composed of the following 4 steps:Firstly,Identify the highlight and shadow pixels,and fill these pixels by approximate nearestneighbor matches.After that,the image only contains the diffuse component.Secondly,Separating shading and reflectance components from the diffuse image.Because of the ambiguity in the decomposition problem,it will fail and generate artifacts in regions that having rich local details of reflectance or shading.We ask users to indicate global shading and reflectance information to prevent them from affecting each other.Thirdly,a joint optimization algorithm is then introduced to recover the normal and illumination variations from the shading map.Finally,we assign the specular parameters to each pixel guided by user strokes and the reflectance map.In addition,we provide a convenient UI for user interaction,which make it easier to deal with details,and it support real-time view of the intermediate results,so that user can modify the interaction according to the result.Experimental results show that our algorithm works well for the images with complex illumination,it have better detail result both in the geometry modeling and material modeling.