Research On Interactive Editing And Blending Of Visual Media

Visual media is an important carrier for human to get information and understand the world, which includes images, videos, and digital geometry. In the family of visual media, videos and images are widely used and can be obtained very easily, which play a very important role in our daily life. The rapid development of multimedia and Internet technologies has greatly accelerated the acquisition, dissemination and sharing of images and videos, which also brings higher require-ments for editing and processing. However, many of the current methods do not provide intuitive interactions and real-time feedback to efficiently process videos and images, which are featured by large amount of data and complex structure. How to design simple interactions for editing images and videos efficiently, according to their contents and features, has become a hot field in computer graphics and computer vision.We have done a lot of research on the key problems of interactive editing and blending of images and videos. Our work focuses on efficient video object cutout, scene-feature-sensitive image blending, and depth-consistent blending on stereoscopic images. The contributions of this paper can be listed as follows:[1] Efficient video object cutout. We propose a novel interface for interactive video cutout, and users can select dynamic video objects step-by-step by dragging a brush when the video is playing. To provide cutout results instantly and correctly, an efficient algorithm based on3D Graph-cut is designed. In addition, a filtering-based temporal-coherent matting is proposed to improve the results of video cutout. Experiments and user study show that our method can help users cutout video objects efficiently and correctly, which also enables many applications of video editing.[2] Scene-feature-sensitive image blending. We propose a method for environment-sensitive image blending, which improves the traditional gradient-domain method by considering the global and local features of the target scene, and produces illumination-consistent results. To deal with inconsistent texture and color in the gradient-domain cloning, we further pro-pose blending methods based on hybrid boundary and hybrid gradient, which produce more natural results and avoid artifacts. [3] Depth-consistent blending on stereoscopic images. We propose an efficient method for depth-consistent stereoscopic composition, in which a source2D image is interactively blend-ed into a target stereoscopic image. Users can edit the disparity maps of2D images by strokes, while viewing the stereopsis from the composition results instantly. Experiments and user study show that our method can not only help users interactively produce stereoscopic composition with high sense of reality, but also enables many applications of stereoscopic image editing.