A Method for Filtering Noise Data by Blending Local Least Squares Fitting Curves

In field of numerical analysis, fitting points in 2D plane with a smooth curve is a widely investigated problem. In this paper, we propose a novel fitting method, which has ability of creating smooth curve approximating the points and filtering noises in the data. Our method is constructed based on the idea of blending local least squares fitting curves with radical weight function. The method first generates a polynomial approximation for each point based on least squares method. Then, these polynomial curves are locally blended with appropriate weights. Finally, a smooth curve is generated, which approximates the 2D data as defined by an error metric based on least-squares technique. Experimental results show that our method has a stable performance and can be used to process all kinds of data in different resolutions.