Image-based Light Modification

Designers translate their ideas into images via visualization tools. Most of these tools provide some level of editing capabilities. Although these tools have very powerful and comprehensive capability in editing, they require users to have a strategy before using and decide the combination of commands he/she needs to apply. For some simple task, for instance, moving the position of the light source in an image, these tools will have very limited power (image processing) or require great amount of computational resources (render). Even though the user may have a clear idea of the expected result, current tools will have a hard time to accomplish the task, mostly because they are ‘content-blind’. To test the idea of a ‘content-aware’ image processing tool, in this research I used machine learning methods to develop a proof of concept tool that is able to rapidly relight an image based on content traits in the image (depth, shape, reflectiveness, etc.) that relate to the re-lighting task. By doing this, we can have the light weight convenience of an image processing tool as well as the content-awareness of a rendering tool.