实值纹理自适应描述在图像分类中的应用

采用概率框架下基于实值最优小波包基的纹理自适应描述准确分析纹理,并同最大似然分类方法相结合,对纹理图像进行分类研究。充分发挥自适应优点,提出将各纹理类最优描述进行组合的方法,使分类准确率从91%提高到95%。     

基于双目视觉的三维重建中特征点提取及匹配算法的研究

三维重建是计算机视觉技术的主要内容之一。基于双目视觉的二维图像特征点的提取及特征点的匹配是三维重建技术的核心。通过以双目立体视觉理论为基础,提取立体图像的Harris角点作为特征点,经过基于小波变换的子线段匹配方法,较好地的解决了匹配精度与匹配速度之间的平衡问题。采用该方法对水泥冷却机内熟料的高度进行了测量,实现了熟料的三维重建。实验表明,该方法可提高图像匹配的速度和精度,能够较精确地对物体实现三维重建。     

基于复小波能量谱的蛋白质活性位点识别新方法

蛋白质活性位点的识别对于理解蛋白质的功能及计算机辅助药物设计具有重要的意义. 基于复小波能量谱建立蛋白质活性位点识别新方法, 采用Morlet复小波对数字化的蛋白质序列进行一维连续小波变换. 结果表明, 通过时-频分析, 能量集中区域往往与蛋白质的活性位点具有密切联系, 并且同源蛋白质序列的复小波能量最大值通常分布于相同的频率处, 表明小波功率谱在预测蛋白质活性位点方面具有广阔的应用前景.     

模拟混凝土孔溶液中钢筋电化学噪音测量及小波分析

电化学噪音(Electrochemical Noise,ECN)测量可用于在无电信号扰动条件下检测腐蚀体系的暂态行为,获得有关腐蚀类型和腐蚀速率的信息.小波分析不需要对ECN作稳态假设,并同时具有时间分辨和频率分辨的特点,在ECN信号处理中表现出一定的优势.本工作考察了氯离子对钢筋在模拟混凝土孔溶液中电化学噪音的影响,并采用离散小波变换(DWT)及能量分布图(EDP)分析ECN信号的时间～频率特征.结果表明,在含NaCl0.0001mol/L的饱和Ca(OH)2溶液中,时间常数为16～32s的暂态占优势;在Cl-浓度更高的溶液中,去钝化趋势为主导事件,表明钢筋在SPS溶液中活化/钝化的临界Cl-浓度介于10-4～10-3mol/L.     

基于小波分解和进化策略的图像融合方法

提出了一种基于小波分解和进化策略相结合的多聚焦图像融合方法。该方法首先对不同聚焦点图像进行无下采小波分解，并分别将其各方向、各尺度的高频信息进行叠加，然后依据它们在高频信息叠加层上对应的局部区域能量的差异，采用进化策略进行区域划分，进而实现分区域融合。实验结果表明所提出的方法比小波变换法具有更好的融合效果。     

Measurement of out-of-plane static and dynamic deformations by processing digital speckle pattern interferometry fringes using wavelet transform

In this paper, we are presenting measurement of static and dynamic deformations in rectangular plate fixed at one end using digital speckle pattern interferometry (DSPI). To improve the measurement accuracy, we have developed a new filtering scheme based on combination of average/median filtering and Symlet wavelet filtering which enhances the signal-to-noise ratio (SNR) in the speckle interferogram obtained from DSPI. Experimental results show that our filtering scheme is quite effective in improving SNR of the speckle interferogram. The measurements by DSPI and calculations from beam theory in the case of out-of-plane static deformation and the measurements by DSPI and accelerometer in the case of dynamic deformation are in good agreement.     

用小波锥分解提取旋转和尺度不变特征

利用极坐标下Fourier变换的模所具有的旋转不变性以及Mellin变换的模所具有的尺度不变性,经过小波锥分解,提取其高层低频分量作为特征进行模式识别。对于0～9的十个不同旋转角度和尺度大小的160幅图像都能正确的进行识别。     

Admissible vectors for the regular representation

It is well known that for irreducible, square-integrable representations of a locally compact group, there exist so-called admissible vectors which allow the construction of generalized continuous wavelet transforms. In this paper we discuss when the irreducibility requirement can be dropped, using a connection between generalized wavelet transforms and Plancherel theory. For unimodular groups with type I regular representation, the existence of admissible vectors is equivalent to a finite measure condition. The main result of this paper states that this restriction disappears in the nonunimodular case: Given a nondiscrete, second countable group with type I regular representation , we show that itself (and hence every subrepresentation thereof) has an admissible vector in the sense of wavelet theory iff is nonunimodular.

     

Spectra of subdivision operators

Let be a sequence of complex numbers and except for finitely many . The subdivision operator associated with is the bi-infinite matrix . This operator plays an important role in wavelet analysis and subdivision algorithms. As the adjoint it is closely related to the well-known transfer operators (also called Ruelle operator).

In this paper we show that for any , the spectrum of in is always a closed disc centered at the origin. Moreover, except for finitely many points, all the points in the open disc of the spectrum lie in the residual spectrum.

     

2nd-order spline wavelet convolution method in resolving chemical overlapped peaks

A 2nd-order spline wavelet convolution method in resolving overlapped peaks is developed. It determines the number of peaks, peak positions and width through wavelet's convolution, then uses spline function to construct the resoluter, which is used to resolve overlapped peaks. Theoretical proof is given, and the selections of wavelets and parameters are discussed. It is proven that baseline separation can be achieved after processed, the relative errors of peak position and area are less than 0.2% and 4.0% respectively. It can be directly applied to seriously overlapped signals, noisy signals and multi-component signals, and the results are satisfactory. It is a novel effective method for resolution.