Uncertain data exist in many application fields, and there are numerous recent efforts in processing uncertain data to get more reliable results, especially uncertainty processing in clustering method. However, it is one of the urgent challenges to discover clusters with specific shape features. So we present a clustering method for data with uncertainties, and it is called multivariate wavelet principal component analysis-improved clustering using references and density (MWPCA-ICURD), which utilizes feature extraction and density-based clustering. To cluster uncertain data with specific shape original features, the original features are extracted by MWPCA method, which combines digital wavelet decomposition and principal component analysis organically. Then, a density-based clustering method ICURD is explored to discover specific shape clusters. Experimental results illustrate its validation and feasibility.