This paper presents a novel noise reduction technique for images enhanced using random spray sampling methods. The technique uses a dual-tree complex wavelet transform to analyze the luma channel of the original non-enhanced image and enhanced noisy image. The standard deviation of wavelet coefficients in the original image is used to shrink coefficients in the noisy image. A noise-reduced version is then computed by mixing the shrunk coefficients with those from the original image based on directionality. Evaluation of results confirmed the validity of the proposed partial-reference approach for noise reduction without assumptions about the noise characteristics.