This is possible with some amount of custom code.
To implement this on server side, you should develop custom frames interceptor Java class. In this class, you can handle every frame picture in YUV format pixel by pixel. Note that this approach increases server load (at least 1 CPU core to encode 2 720p streams), because every frame in a stream should be decoded and then encoded again.
To implement this on browser side, you can use third party library to replace background in browser, then draw result on canvas to capture it and publish to WCS server. Please read this article about canvas streaming. Note that this will not probably work in mobile browsers.