Shape from intensity ranges
Shape from intensity ranges(2011)
摘要
There is a large body of work done in the field of Shape from Shading (SFS). Each of the existing methods however has limitations. Furthermore, the dominant SFS-based 3D feature recovery algorithms were created and tested using images obtained in laboratory conditions, with a known point source light ideally at an infinite distance and objects with Lambertian surfaces. Despite their inherent limitations, these methods are still often used for the shape recovery of smooth, featureless surfaces. This thesis describes a new method for analyzing images, Shape from Intensity Regions (SFI) and a new Shape from Shading method, Direct Shape from Intensity Regions (DSI). SFI is a new image segmentation method grouping together similar isophote region border curves. SFI works with multiple and different types of light sources, objects with textures and multiple albedos, and with objects with non-Lambertian surfaces. SFI is based: on a novel analysis at the isophote region level; new more consistent algorithms for calculating image gradient; new methods for recovery and representation of different lighting scenarios, surface shapes, textures and material properties. The DSI method uses the image intensity regions for recovering the object surface normal by propagation. The propagation can start either from occluding contours or singular points. Direct Shape from Intensity Regions explicitly addresses brightness quantization errors, which can affect the performance of traditional Shape from Shading techniques. The DSI is mainly applicable to smooth diffuse surfaces but is extendable to surfaces with well understood refection models. The accuracy of the DSI technique was measured using synthetic images of simple objects with Lambertian reflectance, as well as real objects of known geometry. The normal map recovery error in these tests is well below seven degrees.
更多查看译文
关键词
new Shape,Shading method,DSI technique,Direct Shape,existing method,Intensity Regions,intensity range,DSI method,new image segmentation method,new method,traditional Shape
AI 理解论文
溯源树
样例
![](https://originalfileserver.aminer.cn/sys/aminer/pubs/mrt_preview.jpeg)
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要