薄板样条matlab代码-CIS-581_Face_Morphing:无缝地将一张脸变形为另一张脸

Overview: ------------------ This project focused on image morphing techniques. The goal of this project was to create a "morph" animation of my face into another person's face. A morph is a simultaneous warp of the image shape and a cross-dissolve of the image colors. In this project, I implement two morphing methods (point triangulation and thin plate spline). The techniques utilized in this project were referenced from the following sources. - https://alliance.seas.upenn.edu/~cis581/wiki/Projects/fall2016/project2/main.pdf - http://cseweb.ucsd.edu/~sjb/eccv_tps.pdf Project Walk-Through and Results: -------------------- **Define Correspondences** - First, pairs of corresponding points are selected by hand across the two input images. This portion of my code leverages MATLAB's built-in "cpselect" tool. <p align="center"> <img src="https://cloud.githubusercontent.com/assets/9031637/20342033/c413059c-abb6-11e6-9ea7-4e75bf347f61.png" width="800"> </p> **Image Morph Via Triangulation** - The point correspondences across two images are averaged, giving an average face. Delauney triangulation is then performed on these points, sectioning the plane into triangles. <p align="center"> <img src="https://cloud.githubusercontent.com/assets/22136934/20359373/89a8e934-abfd-11e6-854d-070eb894a3cb.jpg" width="650"> </p> - These triangles computed in the average face can be applied to the points in each of the original faces. <p align="center"> <img src="https://cloud.githubusercontent.com/assets/22136934/20367661/770dee70-ac1e-11e6-9368-c25225e5082c.jpg" width="650"> </p> - For each pixel in the new image, we determine where that point lies in the two original face images. The new pixel is computed as a weighted average of the original pixels. The key insight here is that points that lie within a triangle in the original face images remain within that tiangle after a transformation into the new image. The output from this step, which is also the final morphed image, can be viewed below. <p align="center"> <img src="https://cloud.githubusercontent.com/assets/22136934/20368113/97a61728-ac20-11e6-86c7-1e7a798cfaad.jpg" width="650"> </p> **Thin Plate Spline (TPS)** - The thin plate spline technique computes an average face as before, by averaging the corresponding points between the input images. For each input image, a function is then computed that exactly maps the selected points in the input image to their corresponding location in the averaged image. This function also applies to all of the unselected pixels in the input image - note that there is no specified "exact" location for these points. Instead, for these remaining points, the function acts to minimize bending contortions to the image. You can see the morphing result from this technique below. <p align="center"> <img src="https://cloud.githubusercontent.com/assets/22136934/20368767/6f9b45b6-ac23-11e6-82b9-5c27dc67eadb.jpg" width="650"> </p>