Non-local means provides a very powerful framework to denoise digital images. Nevertheless, there are several influential parameters on this methodology that are data-dependent and difficult to tune. This paper presents an adaptive image denoising algorithm that uses the non-local means in conjunction with the turbulent particle swarm optimization (i.e. TPSO) which based on a no-reference metric Q. The proposed denoising algorithm can deal with the noisy image even if no "noise-free" reference is available in most practically circumstances. In this paper, we combined TPSO and NLM to propose the TPNLM filter. The proposed filter is able to denoising Gaussian noise without need for any knowledge about the noise-free image, at the same time preserving fine image details, edges and textures well. We also demonstrate several simulations with images contaminated by additive Gaussian noise under unknown noise variance to show that the performance of the proposed method surpasses those of previously published works, both in visual and in terms of peak signal to noise ratio (PSNR) and the structural similarity index (SSIM), respectively.