A method for removing and replacing lighting effects using image sequences
Image sequence analysis has been used extensively for tasks such as the detection of moving objects, tracking, segmentation, automation, and generation of 3D models. There has been much research in the area of normalizing lighting, removal of shadows, and enhancement of images and their contrast. Some image sequence algorithms try to reduce variation in lighting conditions using simple background subtraction, but the use of the content of image sequences to help with the removal and replacement of degraded areas caused by these lighting conditions has not been attempted. This thesis presents a new method promoting automated image reconstruction given a same-scene (from a fixed camera) set of degraded input images. The method determines appropriate technique(s) using reconstruction goals and characteristics of the degraded content, and creates a reconstructed image of the scene for each technique selected. The method ranks the reconstructed images to determine which one is most likely to represent the actual scene. A series of experiments comparing the methods ranked images to best captured images illustrate the strengths of this new method.