Multimedia transmission over the Internet has its quality of service (QoS) requirement. However, Denial-of-Service (DoS) attacks launch large volumes of traffic and consume network bandwidth, thus degrading the quality of the delivered multimedia service. In this paper, we present a differentiated service protection framework consisting of anomaly traffic detection and resource management. Data mining based anomaly traffic detection is implemented at the host; whereas resource management is responsible for the allocation of network resources to the applications. Based on the deviation of the monitored traffic to the normal traffic, differentiated resources are allocated to the traffic. When suspicious traffic is detected, multimedia transmission reacts proactively via adjusting the transmission rate, which can avoid possible packet loss resulting from DoS bandwidth consumption, and thus provide better QoS. The impact of the traffic anomalies on the legitimate services can be mitigated, and the legitimate services can be protected. Simulation results show that better PSNR can be achieved for multimedia transmission during an attack in our proposed framework.