Coherence enhancing diffusion filtering deals with the problems of completion of interrupted line and enhancement of flow-like features such as fingerprints. It is steered by structure tensor which is generally calculated by component-wise convolving between gradient of an image and Gaussian kernel. However, the Gaussian kernel cannot preserve the image structure well. To handle this problem, we propose a novel structure tensor based on connected component analysis (CCA) and apply it to CED filtering. The CCA based structure tensor (CCA-ST) is constructed by combining Gaussian kernel and CCA map. Although CCA is a simple and intuitive method, the experimental results show that CCA-ST provides more faithful results than linear structure tensor.