Recently sparse coding has been successfully used in robust visual tracking. However, sparse coding is very computationally demanding as it needs to solve the L1 norm minimization problem. In this paper we propose a visual tracking algorithm based on both holistic and local appearance modeling. For the local appearance model, we use salient coding instead of the usual sparse coding to encode image patches sampled from each frame. Salient coding method exploits K closest codes to obtain a salient representation, which can be implemented efficiently. In our tracker, we combined the strength of global and local models to form a robust and effective tracking approach. Moreover, we propose a simple yet effective update strategy adapted to our collaborative model to deal with appearance change and reduce the drifting problem. We tested our algorithm on several challenging image sequences involving partial occlusion, drastic illumination change, pose change, and fast motion. Experimental results show that the proposed algorithm performs well against several state-of-the-art methods.