Wavelet transform is a superior multi-resolution analysis method, which has the characteristics of time-frequency localization and multi-resolution, and can perform time-domain and frequency-domain analysis simultaneously. The wavelet transform can be used to finely analyze the image components, and effectively suppress the image noise when enhancing the image contrast, edge and other features.

 

The general steps of image enhancement based on wavelet analysis are as follows:

(1) Perform wavelet transform on the original image to obtain decomposition coefficients of different scales and directions;

(2) According to the purpose of enhancement, calculate the gain coefficient, so as to modify the wavelet decomposition coefficients in different scales and directions, so that some interesting components are amplified and some unnecessary components are reduced:

(3) Based on the modified wavelet decomposition coefficients, the enhanced image is obtained by inverse wavelet transform.

The coefficient modification methods in step (2) mainly include high-frequency enhancement, unsharp mask and sub-band enhancement.

High-frequency enhancement is to increase the high-frequency coefficients as a whole, relatively highlight the high-frequency components of the image, and enhance the edge of the image. However, because the obtained image is often bright or dark, the contrast may become poor. It needs wavelet inverse transformation before contrast enhancement to obtain High quality enhanced images.

The unsharp mask is to blur the image before preprocessing (equivalent to low-pass filtering) and calculate the difference with the original image. The difference image is multiplied by a correction factor and then summed with the original image to achieve the purpose of enhancing the edge of the image. The unsharp mask based on wavelet transform has obvious advantages compared with the traditional method. Firstly, the wavelet transform separates the detail features of different resolutions in the original image with different scales, avoiding the tedious work of continuously adjusting the window of the low-pass filter to select the enhancement effect in the traditional method. Secondly, since the wavelet coefficients at different scales are enhanced separately, no matter the thicker or thinner edges in the original image can be enhanced.