Research On 3D Shape Recovery Based On Single Image

Recovering the 3D shape based on a 2D image is been a hot topic in computer vision.The shape from shading(SFS)is an important reconstruction technology that the 3D shape of an object is recovered from a single image.There is a relationship equation between the shape of 3D surface and the gray value of the image.The depth information of object surface would be obtained by solving the equation.In this thesis,a new optimization algorithm based on the improved Phong illumination model is proposed,which overcomes the shortcomings of existing SFS algorithms.The main research are summarized as follows:(1)Research on illumination model and projection model in SFS model.Because of the limitations of the illumination model used in traditional SFS algorithm,a study on the scope and deficiency of the existing illumination model has been done.In addition,an improvement of Phong illumination model was proposed.The Oren-Nayar reflection model was used to instead of diffuse reflection in the Phong model.It would avoid the error of the original object,which caused by the diffuse reflection of Lambert model.In order to overcome the disadvantage of the orthogonal projection model,the perspective projection model of actual camera imaging relationship was used in this thesis.It would generate a more effective irradiance equation.(2)Numerical algorithm research.A study on the SFS algorithm based on Lambertian model under orthogonal projection and the SFS algorithm under perspective projection has been done.the principle?time complexity and algorithm accuracy of the process for solving the irradiance equation of them was analyzed.According to the low accuracy of those numerical algorithm,the irradiance equation under the Phong model was established and was converted to the Hamilton Jacobi partial differential equations,which contained the depth information.The high-order local Lax-Friedrichs scheme and the improved five-order weighted essentially non-oscillatory scheme was introduced to solving the Hamilton Jacobi equation and improving the algorithm for solving HJ equation,which could effectively improve the reconstruction accuracy of SFS algorithm.In order to verify the effectiveness of the SFS algorithm based on improved Phong model proposed in this thesis.The single synthetic image and the single real image was reconstructed.The reconstruction results was compared with the result of existing algorithmqualitatively and quantitatively.The test showed that the proposed algorithm had a better visual effect.Compared with the orthogonal algorithm,the average error was decreased by62.2%,compared with the similar algorithm,the average error was decreased by about 30.7%.