Adaptive filtering for medical image based on 3-order tensor field

Structure tensors are a common tool for orientation estimation in image processing and computer vision. But the problem is that there is more than one significant orientation in corners, junctions, and multichannel images in second order model, In this paper, We provide a theoretical analysis for high order structure tensor which contains more configuration information in corners, junctions, and multichannel images better than second-order model. This paper deals with the problem of de-noising using 3D statistics. We present a novel non-iterative filter algorithm based on 3D tensor field. The algorithm consist of the following steps. Firstly, we sample every simple neighbourhood of the 3D data set and produced 3D tensor field according to a certain mapping relation. Secondly, According the characteristic of tensor filed, we can externalize the local orientation information which can control the size, shape and orientation of the filter. The high- pass filter along with the main orientation of tensor to protect structure information from damage and low-pass filter along with the weak orientation denoising. We have tested this algorithm by medical image. The PSNR increased 32% averagely and the structure information was reserved well. Nevertheless, Better results the three order tensor got at the cost of low operating speed. Further work is to develop the operating speed in three order tensor.