High-dimensional clustering arises frequently from many areas in natural sciences, technical disciplines and social medias. In this paper, we consider the problem of binary clustering of high-dimensional data, i.e. classification of a data set into 2 classes. We assume that the correct (or mostly correct) classification of a small portion of the given data is known. Based on such partial classification, we design optimization models that complete the clustering of the entire data set using the recently introduced tight wavelet frames on graphs.1 Numerical experiments of the proposed models applied to some real data sets are conducted. In particular, the performance of the models on some very high-dimensional data sets are examined; and combinations of the models with some existing dimension reduction techniques are also considered.