Content-Aware Image Editor
Open the editor →An implementation of seam carving, from the Avidan and Shamir paper. I wrote the first version for Computational Photography in the OMSCS. Instead of scaling an image, it removes the lowest-energy connected path of pixels one at a time, so detailed areas keep their shape and flat areas lose pixels instead. Plain gradient energy leaves visible artifacts on complex images, so seams are picked by forward energy, and a saliency map keeps them off faces. Masking a region and forcing seams through it gives you object removal. The first version was slow. I vectorized the energy and dynamic-programming passes and recomputed energy only near each removed seam, which cut runtime by about 10x.