A Study Of Image Forground Exaction Based On Defocus Map

Foreground extraction, is the task to extract the foreground region, which may be in arbitrary shapes, and attract most visual attention in images. As one of the basical operations in image processing and the most commonly used and most important operation in visual effect generating, foreground extraction has been widely used in the fields of magazines, two-dimensional photography arts, television programme, advertising etc.The existing algorithms used for foreground extraction can be mainly classified into two categories:image segmentation and foregound matting according to the assumption of whether there arc semi-transparent pixel. In this paper, after a brief introduction of the research background of foreground extraction, we summarized the recent development and analyzed the pros and cons of a series of related algorithms, most of the algorithms need lots of manual work.With the rapid development of the DSLR and photography enthusiasts, there are more and more professional looking photos on the internet. We can find that photographs are different from the normal vision of humans in the form, since the optical properties of the camera lens are used to create a center of interest in the image. The center always attracts people’s attention. Photographers let the feelings of the eye and lens performance exposure in the image seemlessly using the combination of lens and eyes. And their own area of interest can be passed to the viewer.When people are introducing SLR camera or other shooting equipment, they will always emphasize the superb performance of bokeh. Bokeh, to make it simple, is the foreground-clear-background-blurred effect between the object and the background. This kind of effect makes the object stereoscopic, emphasizes the object and distinguishes the area between the foreground and the background to a certain degree. Bokeh can be represented by depth of field in image processing.We first do the image defocus magnification to do the image enhancement just like the images from SLR cameras.Then we give the forground a new definition for the photographers:the region what photographers want to focus on and what is salient in the image at the same time. We use a defocus map which represents the distance of object from camera, and cooperate with saliency based approach. Based on the new definition, we propose a new method to extract the foreground which do not need any user interaction.Specifically, for a given image, we firstly extract a Saliency Map (SM) and a magnificated Defocus Map(DM), with their value corresponding to saliency value and depth of field value. Then we segment images and extract segment feature based on these two extracted maps. A classifier is trained to separate foreground from background. On test images, segments are considered to be foreground if they have high enough confidence on the trained classifier. The proposed method is tested on MSRA Salient Object Detection image set and Flickr image set. Experimental results demonstrates that our method can obtain better performance on foreground extraction over related methods.