Single view modeling of symmetric curved surfaces and texture extraction
单视图对称曲面建模及纹理提取

An entire demonstration on that the horizontally flipped image of a reflective symmetric object was equivalent to an image photographed from another position was given. The parallel structure was used on calibrating the camera first. Then the feature points were matched between the original image and the horizontally flipped image. Traditional structure from motion techniques can be used on recovering these points. With the constraints of these feature points, perspective projection, coplanarity and colinearity, the curved surfaces can be recovered by minimizing the linearly constrained quadric objective function. Finally, the pixels of the recovered texture were obtained as a barycentric combination of the colors of the near four pixels. The effectiveness of this method was verified by experiments.