Automatic Foreground Extraction Of Static Image

This paper presents a algorithm about fully automated matting of digital image,this algorithm is mainly for image which is a large scale and single color background.In the studio, photography studio and some other situations, monochromatic sceneryis often used as background for camera and take pictures, and then extract the interestregion by PS(Photoshop) or other image processing methods, and to merge in thebackground. Such a situation, this paper presents a method of image extractionwithout human interaction that for high-pixel image. In this method, several digitalimage processing algorithms are involved, like: sampling for background, extractionof main area which is interested, contour extraction, the unknown region partition,opacity calculations, Knockout algorithm as matting algorithm, in order to achievebetter results in the process of unknown region differentiating, this article uses thesimple face recognition and edge detection algorithm. The algorithm process isdivided into three parts: automatic region partition, opacity calculations andforeground updating. Algorithm uses the HSV color space in automatic regionpartition, and then extracting the foreground subject area of the image, according tothe contour line of the subject area, the source image could be divided into foregroundregions, background area and unknown area, if the image contain pixels which aretranslucent, regard the area which consist of these pixels as unknown, and mergingthe translucent area into the unknown to complete the Triamp division; In the opacitycalculation process, a improved Knockout matting algorithm is proposed, and presentsa method to extract sample pixels base on the image clustering, using the extractedsample pixels and the fusion equation to calculate the opacity of every pixel inunknown region, and this opacity consist of three components which correspondingthe three channels of the RGB space, then use the weighted method of Knockoutalgorithm to get the opacity of the calculated pixel; In foreground updating process,the result of the foreground extraction is need to be revised, and combine theforeground sample pixels and the background sample pixels one by one to find a pairof foreground and background value, so that they are fused according to the fusionequation and the opacity, record the fused value of C’, and compared with the current value of the source images C, choose the pair of foreground value and backgroundvalue which make the difference between C and C’ get the smallest value as theoptimal solution, so the foreground value and opacity which are extracted by theabove-method are the all information to be get, and the image fusion can beperformed according to this information.The proposed approach is able to handle larger scale image in a short time,namely the real-time performance is good; For small objects, such as hair, fluffyobjects in the images of the human or animal hair and so on, can be extractedeffectively; In terms of transparency calculation, the matting algorithm is used, so theedge extraction is mellow, and the result contain the additional transparency value ofeach pixel, and lay a good foundation for image fusion; As there are translucentobjects in the image, such as glass, glasses, wedding dress and transparent plastic andso on, this paper has a improved algorithm which can effectively completed thecalculation of the translucent area of Alpha (transparency) value, and the algorithm isability to accurately estimate the translucent of the foreground pixels, the extractionresults will not include the background color.