Matting-based Interactive Image Editing

With the development of World Wide Web, It is much easier for users to get varieties of multimedia resources efficiently, especially the images and videos, which have become the important element of internet and have deeply impacted our daily life. The recent image and video processing technique focus on how to edit the multimedia with the specific needs instead of simply store and organize in computer it. Therefore, how to edit images and videos efficiently and conveniently has become an important issue.Normally, we are only interested in a part of object in the image, such as human beings, animals, buildings, etc., which we called foreground, and the other parts are called background. The main task of image extraction is to accurately extract the foregound from complex background with limited interactive operation.The main task of image composition is to paste the foreground into a new background image or image sequences. And adjust its color, brightness, etc. and make the foreground and background integrated together naturally, no artifactcan be noticed.Extraction and composition techniques are very applicative, such as photo editing, which is implemented by photoshop. The technique can also be applied to film production, video conference and pattern recognition.The research on these two areas has important theoretical significance and wide application prospects.The main task of this dissertation is to improve the current image editing techniques. In order to improve the accuracy and efficiency of these techniques, an interactive editing system is designed in this dissertation.In term of extraction, firstly, a texture-sensitive segmentation method is applied to matting technique as a pre-proceeding procedure. Our method can distinguish different textures by measuring the similarity of different regions. Secondly, multi-scale technique is used in the segmentation to improve the results. Finally, trimap is produced by the result of segmentation to achieve a high quality matteIn term of composition, a video composition method via optimized 3D mean value coordinates and hybrid blending is introduced. A boundary-aware algorithm is provided to optimize the 3D mean value coordinates. Moreover, matting-based approaches are used to refine the boundary of foreground. Our method can also be applied to image cloning and object removing.The main contributions of our dissertation are as follows:1. A texture-sensitive segmentation method is proposed to improve the foreground matte in high-texture region.2. A video composition method via optimized 3D mean value coordinates is introduced in video composition. A boundary-aware algorithm is provided to optimize the effect of 3D mean value coordinates.3. A hybrid composition method is introduced in video composition it can improve the effect of composition boundary.