Project:Gradient Domain Fusion

Toy Problem

This project mainly explores the gradient domain fusion. To demostrate the correctness of algorithm, a toy problem is implemented. The result is shown as follows. From the result, we could see that it is same with the original image.

Source Figure	Result Figure

Poisson Blending

Three experiments have been done to demostrate the effect of poisson blending. The results are shown in the following table. Before blending, one mask which defines the selected region from the source image needs to be drawn. Further, the source image region needs to be calibrated and projected to the target image via affine transformation, where rotation, translation and scale can be defined.

Source Figure	Target Figure	Result Figure

What I have learned

In this project, we have learned poisson blending method which is a seamless method for image copy and pasting. Least square is a optimal to solve poisson blending. For each pixel, one 4 connnected neighbour pixels are used, so parse matrix can be used to both reduce the memory complexity and time complexity.