An Advanced Method to Detect and Identify Seam Carving in Digital Images

Seam carving is a content-aware image retargeting method. The seam carving algorithm assigns a Sobel-operator-based energy value to each pixel. Seams are defined as the eight-connected paths of pixels, either vertically from top to bottom or horizontally from left to right. Successive removal of the optimal seams, i.e., those seams with the lowest sum of energy, allows reduction in image size. Pixels with lower energy are generally removed earlier. This basic idea of seam carving makes the modifications of the image difficult to identify. Moreover, we can deliberately assign low energy to particular objects so that they can be removed from the image. Development of a seam carving detection method is therefore very important, although challenging. In our previous granted project, we have proposed a novel method, referred to as “patch analysis“, for detecting seam-carved images. This method uses pipelined mechanisms serving for image segmentation, optimal patch decision and transition probability calculation respectively. A support vector machine is finally trained to detect seam-carved images and identify the hot regions that frequently passed by the carved seams. Experimental results reveal that our method is the most accurate seam carving detection method in the current stage. The achievements have been presented in ACM SIGGRAPH 2013 conference and also accepted by Pattern Recognition Letters. This research project aimed to improve the detection of seam-carved images and the location of carved seams. We probed into the seam carving procedure. In doing this, we updated our formula for selecting the optimal patches and thus increased the accuracy for detecting seam-carved images. We also developed a novel method to identify the hot regions nearby carved seams. The proposed methods gained more outstanding achievements then our previous work.

Project ID:PB10308-3322
External Project ID:MOST103-2221-E182-049