Location fingerprinting is a common approach to indoor localization. For good accuracy, the training set of sample fingerprints, each mapping a fingerprint to a location, should be sufficiently large to be well-representative of the environment in terms of both spatial coverage and temporal coverage. Unfortunately, the task of collecting these samples can be tedious and labor-intensive because one must label each location that is being surveyed. On the other hand, fingerprints without location information are abundant and can easily be collected and so recent studies have tried to capitalize on these unlabeled fingerprints to improve the training set. The paper investigates how this goal can be achieved via graph regularization based on Total Variation (TV). TV is highly effective for semi-supervised learning in image processing but it is not clear whether its success can be transferred to indoor location fingerprinting.