Ceometry Features Reconstruction Of Measured Material Based On Photometric Stereo

The applications about photorealistic rendering technology become more and more widely, and realistic rendering of complex material plays an important role in the field of film and television animation, advertising, computer aided design of product and garment, which is studied and paid close attention to the academic and application field.In computer graphics, the difficulty of rendering complex materials such as fabric and skin arises from that high sense of reality about material representation depends on the geometry modeling and material modeling. Geometric modeling describes the shape features of the object, which is macro-scale geometry to build visible bump features on the surface of object. Material modeling describes the light reflection property of the object, which is micro-scale geometry constructed to influence the distribution of high light. Different objects of the same geometric model will show complex optical phenomena due to difference among the surface geometric structure, material, density, color and other attributes. It will be a good option for Bidirectional Texture Function (BTF) to combine geometric modeling and material modeling.BTF defines and transfers the uniqueness of any surface of material. BTF is able to capture shadow, mask and complex non diffuse reflection effect, which makes the material have a sense of reality with no artificial trace, considering the influence of light source and factors of view. BTF is a six dimensional texture representation method, including light source, point of view and position information, which collects, measures and captures the information of various materials in real world, and expresses the realistic material as a 2D texture under the conditions of transforming the illumination and view. Acquisition system captures the image sequence of material in different light conditions and from the perspective of various directions, which makes up BTF data. In brief, BTF can be seen as a combination of meso geometry structure and reflectance behavior. The article will reconstruct the meso geometry from BTF data of materials, including the normal and depth information, and the reconstruct method is for the surface of measured object, such as sweater fabric, wallpaper, curtains and other flat materials.To reconstruct the geometric characteristic from the image sequence of material,3D reconstruction method in computer vision is often adapted, including recovering the surface orientation, and fitting the surface according to the measuring gradient field. There are many common 3D reconstruction methods, such as Shape From X method, photometric stereo method, the motion image sequence method, stereo vision method, etc. Among them, the photometric stereo method is simple and applicable to 3D reconstruction of any approximate diffuse surface, which is able to reconstruct the normal information of measured material from the BTF data, restoring the details of the scene. But due to the fact that the reflection model is assumed to be ideal Lambert reflection model, and there are non-lambert corruptions in the actual input image sequence such as specularities, shadows, image noise, etc. As a consequence, the results of the surface reconstruction is depart from reality, with low frequency noise. Meanwhile, photometric stereo is strict for the acquisition equipment to introduce systematic error.To solve the above problems, this article proposed an optimization method to reconstruct geometry characteristic of flat material from measured BTF data based on photometric stereo method. At the same time, it will also remove the non-lambert corruptions including specularities, shadows and image noise who effect on photometric stereo method considering BTF data of measured material under different light and view directions. First, reconstruct the normal information, and then calculate gradient filed of the material surface, finally, build a new discrete surface, and make its gradient field equal to the measured gradient field in the global vertical using least squares surface reconstruction method.