Seamlets: Content-aware nonlinear wavelet transform

Abstract

With the rise of mobile media devices, resizing an image or video to fit a screen of arbitrary size has become an important topic. In general, arbitrary resizing does not preserve the original image aspect ratio, and hence, can introduce significant visual distortion. Content-aware retargeting methods have been proposed to preserve salient features of an image or video ([1][3][5][6]) even under arbitrary changing of the aspect ratio. While these are useful tools, little attention has been given to the development of an efficient retargeting-driven representation of images. Such representation, for example, can lead to efficient distribution of media destined for retargeting at a variety of devices with different aspect ratios. We propose a new framework, seamlets, which utilizes the discrete wavelet transform (DWT) to perform content-aware arbitrary media resizing. In essence, seamlets provide an efficient multi-resolution representation for retargeting applications. The seamlet transform elegantly generalizes the DWT and seam carving, which is an intriguing retargeting method. The result is a new image framework, seamlets, that inherits the benefits of retargeting and wavelets.