Image smoothing algorithms based on PDE and variational formulation minimization can denoise effectively while preserving the edge information

but the texture information can not be kept as the noise and texture are difficult to discriminate. A new texture preserving filter is proposed. The new algorithm first uses image decomposition model to separate image into the geometry part and the noise/texture part

then calculates the local variance of the latter which is incorporated with the traditional deviation cost to form a spatially adaptive constraint. The new filter smoothes less in the texture regions

thus could preserve the texture information. The visual effect and data analysis of the experiment result proves the effectiveness of the algorithm.