Bad result when enlarge the quads number or using other test images.

[Hanqer]

When I test your code with other images, if the parameter lambda_l = 100, I will get a bad result image.But if I ignored the line preservation and set lambda_l = 0, the result will be better. Do you face the same situation? Is something wrong in line preservation?

[yuzlee]

There were some known problems, I am sorry that I had not mentioned it.

1. The seam carving uses a MAX_ENERGY to fill the seams, so the
   image would be over-carved if one side has a large blank area (eg. about half of the width/height)
2. As you see, the optimization is sensitive, there are many factors could influence the result, such as the lambda, the size of quad and the scale of the image. The linear system is ill-conditioned in my experiments.

But now I have no time to solve it, so if you could find an available solution, please tell me, or submit a pull request directly.

Thank you!