Non-local Haar transform image denoising method

Abstract

The invention discloses a non-local Haar transform image denoising method. The method comprises the following steps of S1, executing block matching and column scanning row matching on an original noisy image; s2, performing Haar transform on the result, shrinking the transform coefficient by adopting a double-hard threshold, performing Haar inverse transform on the result, and aggregating the image blocks to obtain a preliminary noise reduction image; and S3, synchronously executing block matching and line matching on the original noisy image and the preliminary noise reduction image, executing Haar transform to obtain two groups of transform coefficients, then performing Wiener filtering and denoising, executing inverse Haar transform on the de-noised transform coefficients, and executingimage block aggregation to obtain a final de-noised image. According to the method, simple Haar transform is performed on the most similar pixel group to realize effective image denoising, so that image details can be well reserved, and false signals are rarely introduced; the obtained de-noising result image is greatly improved in the aspect of objective evaluation indexes compared with an existing algorithm, and is superior to the existing algorithm in the aspect of subjective vision.

Description

technical field [0001] The present invention relates to the technical field of image denoising, and more specifically relates to a non-local Haar transform image denoising method. Background technique [0002] Image denoising has always been an important basic research topic. Although image denoising has been studied for decades, image denoising is still a relatively active research direction until today. This is because no matter what the environment, no matter how advanced the equipment is, the image acquired will inevitably introduce various noises. The existence of noise will affect the performance of computer vision, pattern recognition, machine learning and even artificial intelligence. Research brings more or less influence, so people are always pursuing more perfect image denoising results. [0003] Image denoising research is generally divided into spatial domain methods and transform domain methods. These two methods always complement each other throughout the res...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to solve the deficiencies in the prior art, thereby providing a simple weighted average of similar pixels, so that the image details can be well preserved; the proposed method does not perform a large number of complex wavelet transforms like BM3D Discrete cosine transform causes a large number of false signals to be introduced while denoising, and the present invention only performs a simple Haar transform on clustered similar pixels to achieve a much better image denoising result than BM3D

Images