Multimedia data analysis on a massively distributed parallelization network of Anonymous Web Clients

This paper proposes an architecture for forming a parallelization network composed of anonymous web clients - those users anonymously browsing a website. This architecture is layered on conventional HTTP-based web applications to organize their clients as computing nodes in a massively distributed computation. The key technology of this architecture is a dynamic installation mechanism for a media analysis module, and a dual-channel communications protocol established between Anonymous Web Clients and servers in order to distribute work, and to retrieve the computing results. The important feature of this architecture is that the clients are kept anonymous to the server, so the server does not have to keep track of clients' status. This stateless communication architecture enables scalable expansion of the parallelization network's computing power without relying upon a centralized data center. In effect, this allows a web-based video sharing service to offload some of its analysis onto users browsing the site, in the background, without interrupting a user's browsing experience. This paper also demonstrates an important optimization for performing media analysis within the parallelization network, which takes advantage of video inter and intra-frame pixel color homogeneity. This paper shows several experimental results for clarifying the system's feasibility and effectiveness, by using a prototype system implementation.