Seam Carving For Feature-aware Image Resizing

Image resizing plays an important role in digital image processing. It is widely used in various applications including image resolution adjustment, format conversion, and displaying high resolution image on the screen of PDA.Interpolation techniques are generally used to resize image. Unfortunately, these techniques introduce much distortion in the resulted image and there are artifacts in the result such as zigzag effects and blurring near sharp edges. Recently, Avidan and Shamir propose an algorithm called Seam carving image resizing. It repeatedly removes or inserts horizontal and vertical seams to change the size of image with preserving the important contents in the image. However, it can not preserve the sharp features along the silhouettes of important objects in the image because the high level perceptual information is not considered in the algorithm. In order to preserve the features of the image, we propose two image resizing algorithms based on seam carving scheme.First, we propose a local adjustment technique to preserve the important features in the resized image. As the distortion is caused by the seams across the sharp feature of the image, the algorithm adjust the local distorted region of the seams to reduce the distortion. We compute all the seams that will be removed or inserted. If there are seams crossing the sharp features of the image, we classify the intersecting points into different clusters. For each cluster we adjust the image locally according to the slope and connectivity information to preserve the features.We also present an energy optimization algorithm by introducing the energy function that represents the similarity of the feature point's neighbourhood based on the observation that one seam only influences the neighbourhood of the point of intersection.When a seam intersects the features of the image, we can use dynamic programming algorithm to find the best adjusting path to get the most similar neighbourhood after adjustment. So the features of the image will not suffer distortion and the resized image preserves the features of the origin image.