| Digital photography is becoming extremely common in our daily life. However, images are difficult to edit and interact with. From a user's perspective, it is important to interact freely with the images on his/her smartphone or ipad. In this thesis we develop several image editing and interaction systems with this idea in mind. We aim for creating visual models with pre-computed internal structures such that interaction is readily supported. We demonstrate that such interactable models, driven by a user's hand, can render powerful visual expressiveness, and make static pixel arrays much more fun to play with.;The first system harnesses the editing power of vector graphics. We convert raster images into a vector representation using Loop's subdivision surfaces. An image is represented by a multi-resolution feature-preserving sparse control mesh, with which image editing can be done at semantic level. A user can easily put a smile on a face image, or adjust the level of scene abstractness through a simple slider. The second system allows one to insert an object from image into a new scene. The key is to correct the shading on the object such that it goes consistently with the scene. Unlike traditional approach, we use a simple shape to capture gross shading effects and a set of shading detail images to account for visual complexities. The high-frequency nature of these detail images allows a moderate range of interactive composition effects without causing alarming visual artifacts. The third system is on video clips instead of a single image. We proposed a fully automated algorithm to create video loops from short (5-second) videos. We then introduce a novel media format called progressively dynamic video, which encodes a wide spectrum of looping videos, ranging from a static scene to a highly animated video. A user can adjust the level of dynamism in a scene, or locally adjust motion configuration, based on personal taste or mood. |
