Tagging and tracking protein compounds/compounds are key to a better understanding of proteomics such as protein-protein interaction and protein signaling pathway. In this paper, a generalized region tracking framework by statistical particle filter (PF) is presented for tracing the movement of protein compounds in confocal microscopy images. To effectively select the features to be tracked, a grid-based minimum variance spatial sampling method is developed. A similarity distance function is presented for feature correspondence finding. The experimental results demonstrate the tracking performance of the proposed framework for small size protein objects with irregular motions and large shape deformation in highly cluttered environment.