Image processing (i) image deformation (1) image processing of-siggraph 2014_ for rectangular panoramic image restoration

Recently found, read the article, not a few days to forget, so began to write something to record the algorithm that has been studied. Nonsense not much said, today read this document "Rectangling panoramic Images via warping", so do a record.

This literature algorithm is divided into two steps:

1, the first step needs to calculate the image of the minimum energy line, the image of the pre-degeneration.

Read this article first to understand seam carving, the algorithm author has the source code, you can download it to read the algorithm principle. This step allows the image to be magnified to a rectangular image, which is called local deformation. In the process of each iteration of this step, each pixel on the source image may move the position, assuming that the source pixel I (x,y) of the input image is amplified by the seam carving algorithm, and the position moves to I ' (x,y), then we can calculate its field (optical flow field), U (x,y ) =i ' (x,y)-I (x,y). The first step of the algorithm is to calculate the U (x,y) motion vector of each pixel point after the input image becomes a local warping image.

2, the second step to the local warping image rectangle grid Division

After the record is divided, the coordinates of each lattice vertex (x,y), OK, through the first step we know that each pixel through the U (x,y) stadium, after the deformation of the local warping image, and then we put the image of the local warping through

-U (x,y) Sports ground, deformation back, at the same time, the new position V ' (x,y) for each vertex of X,y is recorded, in fact, this step does not need to deform the data of the image, only the grid grid must be directly deformed according to-U (x,y), record the new position of each mesh vertex.

3, then we take these mesh vertex new position V ' (x,y) as the control vertex, using the transformation method of the similar transform, the initial input image is transformed globally. The control vertex of the deformation is V ' (x,y), and the position after the deformation is V (x,y), thus the global image deformation result can be obtained.

Learning this document requires a good study: "Seam carving for Content-aware Image resizing" and "as-rigid-as-possible_shape_manipulation" These two documents can be downloaded to the author of the source code, so it is easier to learn, to fix these two paper, this document is much simpler. Seam carving for Content-aware Image resizing algorithm can refer to my another article: http://blog.csdn.net/hjimce/article/details/44916869 This article address: http://blog.csdn.net/hjimce/article/details/44904271 Author: hjimce Contact qq:1393852684
More resources please pay attention to my blog: http://blog.csdn.net/hjimce original article, copyright, reprint please indicate the source