Image noise level estimation method on the basis of multi-level DCT (Discrete Cosine Transform) coefficients

Abstract

The present invention belongs to the field of the electronic signal processing, and discloses an image noise level estimation method on the basis of multi-level DCT (Discrete Cosine Transform) coefficients. The image noise level estimation method on the basis of the multi-level DCT coefficients comprises: dividing a target image with noise to be estimated into M*N 8*8 blocks, and performing DCT of each 8*8 block; measuring an edge degree of each 8*8 image block; selecting minimum element values, from the set of the obtained edge degrees, contributing 10% of the number of the elements in the set; searching out similar image blocks corresponding to a plurality of image blocks with minimum edge degrees; solving coefficient intermediate values, closely related to the noise level, of the selected similar image blocks; performing following operations on all the 8*8 image blocks to calculate and obtain an image structure correction factor; and performing the noise level estimation according to a non linear formula. The image noise level estimation method on the basis of multi-level DCT coefficients is able to improve the precision and the stability of the noise level estimation and satisfy the real time requirement of the noise level estimation, and has simple hardware realizability.

Description

technical field [0001] The invention belongs to the field of electronic signal processing, in particular to an image noise level estimation method based on multi-level DCT coefficients (DiscreteCosineTransform, DCT for short). Background technique [0002] Images are the most important way for humans to obtain information. According to statistics, 80% of the information obtained by humans comes from images. Image or video equipment will generate various noises during the imaging process, making the image more or less noisy. Taking the imaging process of a CCD camera as an example, photons sense the image sensor, perform photoelectric conversion, and finally form pixel bit values ​​through a series of processing. The main noise sources in this process are: Bayer Pattern interpolation noise (fixedpaterrnnoise), dark Noise (darkcurrentnoise), instant hit noise (shotnoise), amplification noise (amplifiernoise) and truncation noise (quantizationnoise). The existence of these no...

AI Technical Summary

Problems solved by technology

However, the general method has the following deficiencies: first, the accuracy of noise estimation is not high enough, and the stability is not good enough, it often occurs that the estimation is too large when the noise is small, and the estimation is too small when the noise is large; second, some noise levels The estimation algorithm has high computational complexity and is not suitable for fast calculation or hardware implementation

Images