Seam carving algorithm based on gradient vector direction analysis

The direction of gradient vector is a factor being omitted in traditional seam carving. A new seam carving algorithm is proposed in this paper based on gradient vector direction analysis. A low-pass filter is firstly used on the local gradient vector to extract more suitable seams based on the fact that the gradient vector directions in random texture regions are usually disordered. Then two different pixel energy functions are defined, which give different weights to the x, y direction components of pixels gradient vectors based on different path direction of seams. The experiment results show that the method can better protect the image edge and other details and can reach more similar overall visual effect with original images. Quantitative analysis results also show that the algorithm achieves better results in both image completeness distance and image coherence distance than other methods.