Abstract: In consideration that PCA-NLM may easily lead to the loss of image texture in denoising, an improved PCA nonlocal means image denoising method was proposed based on the description of texture features. An image was divided into flat, edge and texture regions based on the description of edge textures of local structure tensor. Besides, PCA dimensions and search areas were adaptively selected according to the characteristic values of the edge textures to improve the denoising effect. The experimental results showed that this method could better retain the details of the image textures in the areas with abundant texture details and was more effective for denoising than that of PCA-NLM