Multi-scale image noise reduction method and device

Abstract

The invention provides a multi-scale image noise reduction method and device. The method comprises the following steps: carrying out three-layer wavelet decomposition on an input N*N image block; performing soft threshold shrinkage on each layer of wavelet high-frequency sub-band; performing bilateral filtering on the low-frequency sub-band after wavelet decomposition; and performing wavelet inverse transformation reconstruction according to the high-frequency sub-band subjected to soft threshold shrinkage and the low-frequency sub-band subjected to bilateral filtering to obtain a reconstructed N*N image block. According to the method, through multi-scale image noise reduction combining wavelet decomposition and bilateral filtering, high-frequency noise and low-frequency noise of differentscales in the Y-channel image can be suppressed at the same time.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a multi-scale image noise reduction method and device. Background technique [0002] After the image is pre-processed by the image signal processor (ISP), it often contains noise signals of different frequencies. In order to reduce the noise level in the image and obtain a higher quality image, Y channel noise reduction is a very important process. . Y channel noise reduction is an important part of image signal processing, and its quality of noise reduction directly reflects the level of the image signal processor. Image noise reduction is usually performed linearly or nonlinearly using neighborhood information in the spatial or transform domain. Linear filtering operation to achieve the goal of improving the image signal quality, making the shape of the image more uniform. [0003] According to the pyramid structure used in image decomposition, multi-scale im...

AI Technical Summary

Problems solved by technology

[0005] After the optical signal is processed by nonlinear systems such as lens, sensor and image signal processor pre-processing, the obtained image signal noise is often not the Gaussian white noise signal assumed by the usual image noise reduction algorithm. Such image signals often have high and low frequency Noise exists at the same time, and the existing multi-scale image noise reduction algorithms all treat the noise signal as Gaussian white noise equally, and use the same filtering process, so the image noise reduction effect cannot achieve the desired effect

In addition, the nonlinearity of the imaging system makes the lens-corrected image continuously enlarged along the radial noise level from the center of the image. It can be seen that the existing image noise reduction processing has the problem of inconsistent radial noise intensity caused by lens correction.

Images