Optimizing Computer Imagery for More Effective Visual Communication

Computers are becoming faster, smaller and more interconnected, creating a shift in their primary function from computation to communication. This trend is exemplified by devices such as cellular phones with cameras, personal digital assistants with video, information displays in automobiles, and computer screens in elevators and refrigerators. As communication devices and viewing situations become more plentiful, we need imagery that facilitates visual communication across a wide range of display devices. In addition, producing effective and expressive visual content currently requires considerable artistic skill and can consume days. The goal of my research is to develop new techniques and user interfaces that enhance visual communication, while making it fast and easy to generate compelling content. My research combines concepts and methods from visual art, perceptual psychology, information processing, and cognitive science to facilitate users in creating, understanding and interpreting computer imagery. I am currently conducting research in three areas: • Image Retargeting, to provide effective small images by preserving the recognizability of important image features during downsizing. • Semanticons, an information visualization method for automatically generating semantically enhanced file icons in desktop applications. • Animation Retargeting, to enable animation created for the World Wide Web to accommodate the aspect ratios of hand-held display devices. 1. Image Retargeting The motivation for this work is the need for tools that allow us to author imagery once, and then automatically retarget that imagery for a variety of display devices as needed. The problem is to retarget an image to a new size and/or aspect ratio in a manner that preserves the recognizability of the important features of the image. To do this, the features of the image must be identified, their importance determined, and then the image needs to be rearranged such that the important features are wellrepresented. Given a source image and a specification for the size of the output image, the retargeting algorithm proceeds as follows, and is illustrated in Figure 4. First, separate the image into regions by segmenting it. This process involves image segmentation and adjacent region combination based on spatial distributions of color and intensity. In order to identify important regions, generate an importance map of the source image based on a computational low level vision model and a face detection method. If the specified size contains all the important regions, the source image can be cropped. Otherwise, the important regions are removed from the image, and the resulting “holes” in the background filled. The updated background is then Vidya Setlur Nokia Research Center