The performance of image denoising algorithms generally depends largely on the selection of the parameters. We address the problem of optimizing the denoising parameters to achieve maximum denoising performance. Most existing methods for no-reference denoising parameter optimization either use the estimated image noise or individual no-reference image quality evaluation metrics. In the paper, we introduce a natural image statistics based on the generalized Gaussian distribution and an elastic net regularization regression model. Consequently, we propose their use for performing no-reference parameter optimization in the state-of-the-art BM3D denoising algorithm, adaptively depending on which of the following most widely used full-reference image quality evaluation metrics is optimized: FSIM, SSIM, MS-SSIM, without any knowledge of the noise distribution.