Research Of Anti-aliasing Shift-invariant Contourlet Transform And Its Application On Image Processing

Contourlet transform with the characteristics of multi-resolution, multi-direction,time-frequency localized and anisotropic which was proposed combining with the directionalfiltering is considered to be a multi-scale geometric analysis method. It has been widely used inimage processing with the constantly developing and improving of contourlet transform’s theory.But the defects of spectral aliasing and shift-sensitivity have affected its application prospects. Thispaper relied on the project supported by Natural Science Foundation Research Projects of ZhejiangProvince, trying to find a new contourlet transform method with features of anti-aliasing andshift-invariance, namely, Anti-aliasing Shift-invariance Contourlet Transform (ASCT) andapplying it to image processing. The main contents of this paper are as follows:(1) Study on the construction method of ASCT. We leaded the way in an overview of imagerepresentation methods, indicated the limitations of the traditional image representation methods intwo or higher-dimensional image’s singularity, which revealed that multi-scale geometric analysismethod could be a better way to express or ‘capturing’ image’s singularity features. And then weanalyzed the causes of spectral aliasing and shift-sensitivity in detail and outlined the limitations ofthe relative improved contourlet transform methods. Based on this, we designed two kinds ofprograms to achieve ASCT. Program one: as wavelet transform using cycle spinning to achieveshift-invariance, we combined cycle spinning and anti-aliasing contourlet transform to achieveanti-aliasing contourlet transform with shift-invariance. Program two: we improved the twostructures of contourlet transform by using anti-aliasing tower filter bank for multiscaledecomposition and non-sampling direction filter bank for direction decomposition to achieveASCT. Experimental results showed that the new method achieved by the two programs had betternonlinear approximation performance than conventional improved contourlet transform and couldbetter express the detail information of image edge and texture.(2) Research on infrared image denoising using anti-aliasing contourlet transform based oncycle spinning. We applied the new method got by program one on infrared image denoising inorder to verify the effectiveness and feasibility of our method. We used hard threshold denosing method to process the noisy infrared image transformed coefficients to achieve the purpose ofdenoising. The contrastive experimental results showed that our method was better than othermethods to suppress noise and display the detail information of image edge and texture.(3) Research on infrared and visible cloud images fusion based on ASCT. According to thecharacteristics of infrared and visible cloud images we proposed the infrared and visible cloudimages fusion method based on ASCT. We took the weighted regional energy fusion rule on lowfrequency coefficients and weighted regional variance fusion rule on high frequency coefficients.Experimental results demonstrated that the fused image got by our method had higher resolutionand was more abundant and much clearer in infrared information and texture details.