An Improved pyramid and sampling based image matting technique

Abstract

Image/video matting aims to extract the foreground object from the image. It is widely used technique in various applications like video and image editing that needs background replacement and to distinguish foreground object from the image. The accurate extraction of foreground object is performed using a mask known as alpha matte. The goal of image matting is to generate accurate alpha matte that can separate foreground elements from background. Wide range of matting algorithms and approaches have been proposed in literature, yet their performance reduces in terms of accuracy when images have high texture and complex background. Thus, there is a need of matting technique that can accurately generates alpha mattes of high textured images. In this thesis, an accurate image matting technique is proposed. The proposed technique utilizes the concepts of texture smoothing filters, pyramid decomposition and entropy based weighting. It is capable of generating high quality alpha mattes and accurately extracting the foreground object from highly textured images. The work also includes a time efficient video matting approach to speed up the video matting process. The image matting process can also be used for fusion of multiple images to construct a well information fused image, that can be helpful in many image processing systems. The thesis also includes an image fusion technique with image matting that accurately gathers the focus information from multiple images and fuse them into a single image. To demonstrate the significance of proposed technique, visual and quantitative comparison with other state-of-the-art techniques is presented. Simulation results reveal that the proposed technique is more accurate as compared to state-of-the-art matting techniques.