Image denoising is a technique for reducing noise in an imprecise image scene. Several methods were developed to reduce noise in digital images yet preserving image quality is a tedious task. Various literatures have already suggested a variety of noise reduction techniques. To address the issue of determining the degree of noise, this article proposes a singular value reduction and convolutional neural network-based noise estimate technique. The larger (head) sections of an image’s single values are controlled mostly by the image’s basic structure, whereas the other (tail) portions are influenced primarily by the degree of noise. The tail portions of singular values increase in proportion to the strength of the noise. As a result, singular values should be desirable characteristics for assessing noise intensity. We initially introduce different noises with specified intensities into a batch of noise-free images before selecting a given number of fixed-sized image blocks with the lowest variance from these noisy shots. To train the network, individual values from these blocks were fed into a neural network as input and their corresponding noise deviations as output. In the estimating step, single values from the noise image were used to anticipate the unknown noise intensity. The experimental findings show that the proposed strategy has a lot of promise. A variety of noise types, such as composite noise and Gaussian blur noise, may be accurately and quickly estimated using our method.