An improved Genetic Algorithms-based Seam Carving method

In a previous work, we proposed a new method to retarget images, i.e. to resize an image on both vertical and horizontal orientation, based on Genetic Algorithms called Genetic Seam Carving. The previous work presented a new individual modeling to represent connected pixels paths (seams) that were handled by Genetic Algorithms for image retargeting. This modeling has an important drawback, a fixed base-pixel position, the pivot. This condition is not interesting when the pivot is in a region of interest, such as an object one wants to remain in the image. Thus, our novel proposal presented in this paper aims at solving this issue, which could decrease the retargeting performance so that flexible seams are achieved and evolved. To do so, we also present new genetic operators for out target problem. As expected, our proposal outperforms the Seam Carving and the previous proposal in terms of image quality.