Image compression has been an important research topic for many years. There are many kinds of compression techniques, but most of these do not fully take advantage of the characteristics of a target application. An example is virtual conferencing, one of the most popular applications in multimedia, which demands a huge network bandwidth. The paper presents an application-specific image compression technique focusing on virtual conferencing. The paper proposes a compression technique that compresses the image background and foreground (the face part) with different compression qualities. Compared to standard JPEG compression, our algorithms may reduce half of the image file size with a reasonably clear quality of background. This paper discusses how to divide an image into a background part and a foreground part automatically, and how to design and compare several different algorithms for compressing these two parts with different quality.