Participatory sensing is an emerging paradigm in which citizens voluntarily use their mobile phones to capture and share sensed data from their surrounding environment in order to monitor and analyze some phenomena (e.g., weather, road traffic, pollution, etc.). Participating users can disrupt the system by contributing corrupted, fabricated, or erroneous data. Different reputation systems have been proposed to monitor participants' behavior and to estimate their honesty. There are some attacks that were not considered by the existing reputation systems in the context of participatory sensing applications including corruption, collusion, and on-off attack. In this paper, we propose a more robust and efficient reputation system designed for these applications. Our reputation system incorporates a mechanism to defend against those attacks. Experimental results indicate that our system can accurately estimate the quality of contributions even if collusion is committed. It can tolerate up to 60% of colluding adversaries involved in the sensing campaign. This enables our system to aggregate the data more accurately compared with the state-of-the art. Moreover, the system can detect adversaries even if they launch on-off attack and strategically contribute some good data with high probability (e.g. 0.8).