107 Active Contour Based Seam Carving for Content-Aware Image Resizing Abstract-In this paper, we propose a novel noise robust content-aware image resizing method by the idea of seam carving scheme which attempts to preserve important objects in an image while changing the aspect ratio of it. We estimate the local energy map of image by active contour. Based on the energy map we carve and insert connected path of pixels, seam, for resizing image to the specific size. Conventional seam carving method is unsuccessful to keep important objects in an image when the energy content of the objects are low with regard to their adjacent, or the number of seams removed are very large. Since active contour attempts to detect objects in an image, we use it to segment the image in two parts, one contains prominent objects and the other one includes remaining part of the image which is less important; so we can simply carve or insert seams in the second part. The experimental results show that the proposed method produces higher subjective quality images than conventional seam carving method and recent saliency detection seam carving scheme. I. INTRODUCTION MAGE resizing (enlargement or reduction) is a common operation in digital image processing [1]. It is used whenever one wants to change the image resolution. Users now have easy access to high definition image or video content due to the emergence of mobile multimedia proficient devices and accessibility of networks. Usually, these devices have smaller displays and limited processing abilities. Thus, an efficient method to reduce the resolution of image and video contents before delivery to these devices is desired [2, 3]. For example, it is required on a routine basis in digital photography, multimedia, and electronic publishing [4] for adapting the pixel size to the resolution of an output device (printer or monitor) and for generating preview images, or posting digital pictures on the Web. We know that web developing tools can support dynamic changes in page layouts or text but the images have always been a rigid part of these dynamic web pages [5]. Image resizing techniques like cropping [6] or scaling cannot be an appropriate solution since dimension modifications will finally lead to lot of artifacts because of the image content lost during the adjustments. Protection in images often results in user intervention, thus lead to a way for requirement of automatic methods. Original Image Energy Seams Plot Seams Carving Seams Insertion Function Fig. 1. Image resizing comparison for 80% and 120% of original width between previous methods and our suggested scheme. First row is the results of conventional seam carving method, the results of the saliency detection method [10] are showed in second row and the last row demonstrates our results. Avidan and Shamir [7] have developed an algorithm for content aware image resizing known as seam carving. Seam carving proposes a way of smartly removing the pixels of lesser importance in the image. Seam carving is based on the philosophy that human eye mostly notices only the salient features of an image, so gradual addition or deletion of features that is not salient in an image can be an efficient way for image resizing. Some approaches are introduced for resizing videos [8, 9]. In [8] instead of removing 1D seams from 2D images authors remove 2D seam manifolds from 3D space-time volumes. In [9] an efficient algorithm for video retargeting is introduced. It consists of two stages. First, the frame is analyzed to detect the importance of each region in the frame. Then, a transformation that respects the analysis shrinks less important regions more than important ones. A noise robust resizing technique based on seam carving by Achanta and Süsstrunk in [10] was introduced which assigns higher value to visually prominent regions in a saliency map. They figured global saliency of pixels using intensity as well as color features [11]. Since saliency detection method cannot proficiently identify all borders of objects, in this paper, first we detect objects in a given image with active contour based on techniques of curve evolution [12]. So we can detect objects whose boundaries are not necessarily defined by gradient and reach to a binary image map which shows objects with 255 and the remaining part of image with zero. In this map pixels Zahra Toony*, Somayeh Hesabi** and Mansour Jamzad* *Department of Computer Engineering, Sharif University of Technology, Tehran, Iran **Department of Computer Science, Sharif University of Technology, Tehran, Iran toony@ce.sharif.edu, jamzad@sharif.edu, shesabi@mehr.sharif.edu I