| Summary: | 碩士 === 國立交通大學 === 應用數學系所 === 107 === In this thesis, we will discuss how to project 3D face point clouds into
2D space and conform to a mapping of an affine transformation. In
simple terms, an affine transformation is a composite function of linear
transformation plus translation. In addition, the most important issue
in computer vision is the facial deformation. In this thesis, we will also
explore the application of deformation and combine it with affine transformation.
While projecting the point clouds, at the same time make
a continuous expression of the face. Due to the position of the camera,
the lighting conditions at the time and the overall environmental
sensation, etc., the source image and the target image will have color
differences, and we have also made a corresponding color adjustment for
such a problem. When finally adjust the color, we will find that we can
still see the obvious color lines (color difference) at the boundary of the
face. Hence by using image blending, we can make it fade to achieve
a better visual effect. In the process, due to the fact that the detection
landmarks method is unstable sometimes, the boundary of the face
may be bounced sometime. Here we use the cubic spline interpolation
method to make the face more stable.
|
|---|
