Development of Accurate Image Matting Techniques

Abstract

Accurate extraction of foreground objects from various kinds of images/ video sequences is significant yet challenging task in digital image processing. It is an active research area and has diverse applications that includes image/ video editing, image analysis, object identification, detection, recognition and tracking. Image matting, a type of segmentation, aims to separate the foreground object from still images or video sequences, that is done using a mask known as alpha matte. Contrary to conventional segmentation approaches, matting considers the transparency value (opacity) of foreground object, thus decomposes the image into foreground and background by determining the opacity value of each image pixel. This research is focused on exploring the utilization of image matting in different applications such as de-focused region segmentation and fusion of multi-focused images and developing image/ video matting techniques in several domains that includes extraction of underwater objects from images, target extraction from synthetic aperture radar images, extraction and removal of shadows from images, automatic extraction of aircraft from video sequences and extraction of pharmaceutical tablets while dissolving in water. To assess the performance of the proposed techniques, qualitative as well as quantitative comparison is carried out accompanied by state-of-the-art techniques. The simulation results and analysis reveal the significance of proposed techniques.