Over the past few years, fast and robust trackers based on Kernelized Correlation Filters have shown top notch performance on the Visual Object Tracking challenge. However there is still scope for obtaining higher performance through the use of reasonable approximations that can easily be shown to work through empirical methods. We study some variants derived from the Discriminative Scale Space Tracker and show significant improvement in tracking performance. Our tracker outperforms both fDSST and DSST on the VOT 2016 and 2017 datasets in terms of both Expected Average Overlap (EAO) and Equivalent Filter Operations (EFO). We also demonstrate that the error correcting capability inherent in our method leads to a higher performance on the unsupervised VOT 2016 and 2017 benchmarks.