Video foreground object extraction has many useful applications, such as changing background, duplicating foreground objects in the scene, and in virtual reality. The main task is to create an accurate alpha matte that specifies the foreground objects to be extracted. In this work, a new approach for video foreground object extraction is presented; this approach is called "Video Flash Matting". An intermittent flash is used for creating bright frames that are used for creating the alpha matte of the foreground objects through the video. Bright frames contain lit foreground objects that are illuminated using the intermittent flash; however, the background is still under the same illumination without flash. According to that, foreground objects are extracted from the rest of the video using those bright frames. The results of the proposed approach show that Video Flash Matting extracts foreground objects using good alpha mattes without requiring trimaps. Video Flash Matting handles free camera and dynamic background conditions.
展开▼