AN EFFICIENT IMAGE VISION NOISE ESTIMATION FOR ARTIFICIAL INTELLIGENCE APPLICATIONS

Abstract

As imaging technology advances, the expectations of the quality of images are also increasing. Although the imaging sensor always tries to capture the fine and exact details in an image, it is inherently accompanied by specific amounts of noise. There exist many algorithms to drastically remove this noise but they require an accurate estimate of noise present in the image. Thus, the image noise should be estimated accurately in order to obtain noise-free images for artificial intelligence applications. Most of these techniques rely on the transform or spatial domain data of images. For estimating the level of Gaussian noise in digital pictures, we presented a hybrid Stationery Wavelet Transform (SWT) and edge information removal-based approach. A Sobel edge detector is used to remove from consideration the wavelet coefficients corresponding to the spatial domain edges while calculating the noise estimate. Polynomial regression is used to further boost the accuracy of the suggested procedure. Various photos are used to assess the performance of the suggested method. The suggested approach performs far better than all current state-of-the-art algorithms in benchmark tests across a wide range of noise.