Preserving salience by maintaining perceptual differences for image creation and manipulation

点击翻译按钮获取中文摘要

Images on computer screens or on paper are a large part of our everyday lives. Although the number of images we have access to is growing exponentially, the efficiency, beauty, and distillation of the information we see is not improving at the same rate. In my research, I address how to create or extend the communication power of images by mapping visible changes from one domain to another. There is a precarious mapping between how a scene would be perceived and how an image is perceived. For example, although the human visual system does not have a strong absolute sense of color or intensity, images are stored as a discrete set of color or intensity values at each pixel. The goal of this research is to create perceptually salient images by more directly expressing perceived differences in scenes as displayed differences in images. Perceptual difference preservation is useful for may image generation tasks including automatically creating technical illustrations from 3D models, creating a color to grayscale conversion algorithm that preserves all visible color changes through high-dimensional distance metrics, and for hidden feature finding techniques that identify correspondences and mismatches between X-ray and visible light images of historically significant paintings through segmentation and local gradient statistics.