Improved Adaptive Seam Carving for Image Retargeting

In this work, we propose a novel improved adaptive seam carving technique to image retargeting. Experimental results demonstrate that, when compared with the previous seam carving techniques, the proposed one efficiently prevents distortions in the regions of interest (ROIs) and leads to significantly better result image quality. Two importance factors were introduced into the gradient energy metric with which more flexible control of relative extent of content preservation between different image regions can be achieved. One factor is frequency-tuned saliency map, and the other is Canny edge detector. When computing the gradient magnitude, the Sobel operator was considered. In addition, we search for the least energy seam using bi-directional seam operations. In this way, we delete or insert the seam with the minimum energy in all the possible vertical seams and horizontal seams leaving the least damage to the content of the image.