Noise evaluation method and device based on real noise scene and storable medium

Abstract

The invention discloses a noise evaluation method and device based on a real noise scene and a storable medium, and relates to the technical field of image processing, and the method comprises the steps: S1, obtaining a plurality of clean images of different scenes in an environment with sufficient light, and randomly generating a Gaussian noise map, adding the Gaussian noise map into the clean image and expanding the Gaussian noise map to be the same as the clean image in size to obtain a first noise image; s2, acquiring a plurality of images with noise in the same scene in the S1 and in an environment with insufficient light, processing the images with noise to obtain a noise mean value image, and processing the noise mean value image to obtain a corresponding second noise image. According to the invention, noise of different scenes is classified and evaluated, and the method can be applied to photographing and camera shooting related systems and terminal equipment.

Description

technical field [0001] The present invention relates to the technical field of image processing, and more particularly to a noise evaluation method, device and storage medium based on a real noise scene. Background technique [0002] Currently, image noise refers to the factors in the image that prevent people from comprehending the information received by the human eye. Since the image is inevitably affected by the image sensor or the digital signal processor in the process of shooting and processing, etc., the image content of the video image will be wrong, and the image noise will be formed. [0003] However, the traditional noise evaluation methods mainly include the following: (1) It is difficult to accurately find the flat area of ​​the image by estimating the noise parameters according to the flat area of ​​the image, and the noise in the flat area of ​​the image does not represent the noise of the entire image, and cannot adapt to the real image. In different scenes...

AI Technical Summary

Problems solved by technology

Because the image will inevitably be affected by the image sensor or digital signal processor in the process of shooting and processing, the image content of the video image will be wrong, forming image noise

[0003] However, the traditional noise evaluation methods mainly include the following: (1) It is difficult to accurately find the flat area of ​​the image by estimating noise parameters based on the flat area of ​​the image, and the noise in the flat area of ​​​​the image does not represent the noise of the entire image, which cannot adapt to the real In the scene, the noise level in different areas is inconsistent; (2) The arithmetic mean method based on the filter, the calculation deviation of the method is relatively large for the image with rich details, and the adaptability to the real scene is not enough; (3) Based on Noise estimation of wavelet transform This method counts the noise distribution information of the whole image. The degree of noise estimation will be related to the texture details in the image. In the case of small noise, the estimated noise is too large, which is easy to cause subsequent denoising algorithms. make the image too blurry

Images