Adaptive Directional Lifting-based Wavelet Transform For Image Denoising And Its Implementation

The digital images are usually contaminated by the noise in the process of acquisition and transmission. How to remove the noise from the images is an important topic. In 1994, the wavelet thresholding method for image denoising was firstly proposed by Donoho. In this method, the wavelet transform is applied on the image and the wavelet coefficients are treated with shrinkage. Due to its simplicity and high performance, this method is used widely.However, the traditional 2-D discrete wavelet transform is ill suited to approximate the textures and edges of images. The detail information may be hurt by shrinkage and the denoise image may be blurred. In 2007, Ding proposed a separable adaptive directional lifting (ADL)-based wavelet transform. The ADL can search for the direction of high pixel correlation and perform the lifting operation along this direction; therefore, the energy of the high subband will be less. The ADL was firstly used in image compression and got good results due to its superior property. In this article, we use the ADL transform to remove the noise from the image and keep the image texture; further more, a new technique-adaptive omni-directional lifting (AOL) was proposed based on the ADL. Comparing with ADL, which search the direction between±4 5°, the AOL can search for the direction of the strongest pixel correlation in the whole 2-D space of the image between0°°360°. Due to this property, AOL is good at approximating the texture of the image and the result of image denoising will be better than ADL.Due to separable transform and lifting operation, AOL is simple in structure and easy to achieve the inverse transform. Consequently, it is convenient to implement in hardware. Experimental results show that in the texture area of the denoise image, the proposed method outperforms conventional wavelet transform in PSNR and subjective quality.