Research On Directional Multi-scale Image Analysis And Its Applications

Wavelet transform is widely used in many image processing applications, whichinclude image compression, digital watermarking, image denoising, image analysis,image retrieval, and image registration and image recognition. Although great successhaving been achieved, the discrete wavelet transform has three serious disadvantages:shift sensitivity, poor directionality, and lack of phase information.The purpose of this paper is to explore new signal processing methods to enhance thedirection resolution of wavelet analysis. The main research contents of this paperinclude wavelet analysis theory, multi-scale geometric analysis theory, directional filterbank, Contourlet transform and adaptive directional wavelet transform. Wherein thedeveloped methods, such as multi-scale geometric analysis based algorithms areinvestigated to solve some recently encountered difficulties in the applications oftexture analysis and digital watermarking.Texture is usually related to a repetition of a basic pattern in an image, whose localarrangement is regular while in the mass, irregular. Being hot research in recent years,texture analysis is widely used in medical image processing, satellite remote sensing,industrial surveillance and computer vision. However, it is difficult to analyze thetexture due to the diversity of textures, the changes of rotation and scale. Hence, onefocus of this paper is to explore rotation invariant texture analysis methods using thenew multi-direction and multi-scale analysis theory.Digital watermarking provides an efficient copyright protection solution, which canembed proprietary information and product serial number into image, audio and videofiles while preserves imperceptibility of the embedded information due to theredundancy property of human visual or audio system. Digital watermarking methodscan be grouped into two main parts: spatial domain and transform domain. Among thesemethods, wavelet spread spectrum based watermarking method is the most popular one.However, the wavelet spread spectrum watermarking schemes have poor abilities toresist low pass filtering and geometric attacks, such as scale and rotation alterations. In this paper, a new computational effective watermarking method is proposed to settle thedifficulties, which can not only decrease the search space in watermark detection, butalso increase its robustness to low pass filtering.This paper focuses on the increasing of direction resolution of wavelet transform andexplores new methods to solve difficulties in texture analysis and digital watermarking.The main innovative results in the dissertation are as follows:1. A novel method based on the McClellan transform and the quadratic programmingalgorithm is proposed to design 2D FIR filters. The coefficients of the designed 2DFIR filters are optimized by minimizing the integration of the squared errors alongthe desired contours mapped from cut off frequency points in both pass band andstop band.2. A new adaptive directional wavelet transform decomposition method is proposed toachieve a better direction resolution, which is based on the rotation of textureimages. Compare with the Contourlet transform, the new method provides anon-redundancy representation and holds computational efficiency.3. A new adaptive directional wavelet transform decomposition method based on thescalable pyramid is proposed. Compared to the Contourlet transform and theconventional wavelet transform, the new multi-scale analysis tool can provideimage information along arbitrary direction.4. To alleviate the difficulties caused by rotation changes in texture recognition, anovel rotation invariant method based on multi-direction and multi-scale analysisand support vector machines is proposed. And a novel multi-direction andmulti-scale analysis based rotation invariant image retrieval method is developed.5. A novel digital watermarking scheme is proposed to defeat the low pass filteringand geometric attack, which is based on static wavelet transform and directionproperties of the cover images. The new scheme can provide better visual qualityfor ROI region.In this paper, we explore new multi-direction and multi-scale image analysis tools,which are based on existing multi-scale analysis theory and multi-scale geometricanalysis theory. We exploit some new techniques to alleviate the difficulties in texture analysis and digital watermarking. The new theory results can also be used in otherimage applications, such as image compression, image denoising and imagerecognition.