susuzhang的专栏

这是个很好的学习用VC来处理数字图像的教案，想学的可以参考一下，希望对你们有用。

Automatic texture defect detection is highly important for many fields of visual inspection.We introduce novel, geometrical texture features for this task, which are Euclidean motion invariant and texture adaptive: An algebraic function (rational, Padé, or polynomial) is integrated over intensities in local, circular neighborhoods on the image including an anisotropical texture analysis. Adaptiveness is achieved through the optimization of this feature kernel and further coefficients based on a simplex energy minimization, constrained by a measure of texture discrimination (Fisher criterion). A backpropagation trained, multilayer perceptron network classifies the textures locally. Our approach contains new properties, usually not common in feature theories: Theoretically implicit, multiple invariances and an automatic and specific adaptation of the features to the texture images. Experiments with a fabric data set and Brodatz textures reveal up to 98.6% recognition accuracy.

简述了小波变换的基本思想，着重叙述了使用MATLAB 软件小波工具包中的haar, 小波对原始图像进行分解，并对不同分辨率的子图系数进行相应的运算，从而实现对原始图像的压缩、消噪、增强等处理的过程9 得出了haar, 小波变换用于图像处理具有速度快、处理方便、图像压缩比高、去噪效果好、图像特征保持性好等优点的结论，为进一步研究基于小波变换的图像处理技术提供了一定的依据。