[模式识别(视觉/语音等)] 论文
20世纪90年代，贝尔实验室的Vapnik教授第一次提出支持向量机（Support VectorMachine,SVM）的理论与基本概念。SVM方法一种基于统计学习理论(Statistical Learning Theory)的机器学习方法，它以结构风险最小化原则代替经验风险最小化原则，同时结合了机器学习、统计学习以及神经网络等方法[53]。它能够有效的提高算法的泛化能力，解决了小样本、非线性和维数高等难题，并且能够克服传统神经网络等学习算法中网络结构难以确定、收敛速度慢及训练时需要大量数据样本等不足，具有很好的分类和推广能力。
Support vector machine (2018-03-24, Visual C++, 5044KB, 下载1次)
[模式识别(视觉/语音等)] feature-points-matching
对灰度差绝对平均值算法匹配次数多，不具有旋转不变性等缺点，提出一种新的目标识别方法。匹配准则采用具
有环形结构的子窗口内的像素差加毂和的形式表示，保证了算法具有旋转不变性。对模板图像中的特征点按照匹配准则分
别在目标图像中找到相应的匹配点，从而完成匹配操作，与传统的相关匹配算法相比，大大减少了匹配次数。对于因遮挡而
丢失的特征点，可根据已匹配特征点之问的相对距离来重新确定，从而实现目标识别的功能。仿真实验验证了该算法的有
效性。
A new target recognition method was proposed to improve the rotation invariancc and reduce the
matching frequency for nu“mum absolute difference．The rrIatching furlction was the weighted sumof pIx—
els difference in sub—windows． The subwindaws were ring-shaped，which made the algorithm have the
property of rotation invariance． Feature points matching between trac ki“g window and template could be
completed according to the matching functiom(bmpare to conventional algorithm，the matching frequency
is reduced The Iost feature points in tracking window could be determined according to the positions of
the feature points nlatched．Simulation results prove the validity of the method． (2012-02-01, Visual C++, 187KB, 下载18次)
[模式识别(视觉/语音等)] FaceRecog_src
程序结构
整个工程可以分为3个部分：算法、功能和应用。
算法部分
算法部分目前分为4个模块：人脸对齐、光照归一化、特征提取和选择、子空间降维，每个模块是一个项目，每个项目生成一个dll供功能部分显式调用。
功能部分
功能部分只有一个项目FaceMngr，该部分依赖于算法部分，实现人脸注册、训练、识别、导入/导出等具体功能。该项目生成一个dll供应用部分隐式调用。
应用部分
人脸识别Demo.
另外，工程中还有一个项目tools，实现了一些整个工程都可能用到的函数，大部分与OpenCV有关。该项目生成一个dll供各部分隐式调用。
Program Structure
The whole project can be divided into three parts: the algorithm, function and application.
algorithm part
Algorithm is part of the currently divided into four modules: face alignment, illumination normalization, feature extraction and selection, reduced subspace
Dimension, each module is a project, each project generates a dll explicitly call for the functional part.
Features section
Functional part of only one project FaceMngr, part of the part depends on the algorithm to achieve face up, training,
Recognition, import/export and other specific functions. The project generates a dll for the supply of part of the implicit call.
Application Part
Face Recognition Demo.
In addition, there is a project engineering tools, to achieve a number of projects are likely to use the function, the majority
With OpenCV related. The project generates a dll called implicitly for each part. (2011-05-15, Visual C++, 613KB, 下载92次)
[模式识别(视觉/语音等)] plateVs2003
自已用Vs 2003开发的，车牌识别程序源码，很有用的，实现简单图像处理，包括256色转灰度图、Hough变换、Walsh变换、中值滤波、二值化变换、亮度增减、傅立叶变换、反色、取对数、取指数、图像平移、图像旋转、图像细化、图像缩放、图像镜像、均值滤波、对比度拉伸、拉普拉斯锐化（边缘检测）、方块编码、梯度锐化、灰度均衡、直方图均衡、离散余弦变换、维纳滤波处理、逆滤波处理、阈值变换、高斯平滑。
Their own development with the VC, license plate recognition program source code, very useful, simple image processing, including 256 colors to grayscale, Hough transform, Walsh transform, median filtering, binary conversion, brightness change, Fourier transform , anti-color, taking on the number, taking index, image translation, image rotation, image thinning, image scaling, image mirroring, filtering, contrast stretching, Laplacian sharpening (edge detection), block coding, gradient sharpening, gray balance, histogram equalization, discrete cosine transform, Wiener filtering, inverse filtering, threshold transform, Gaussian smoothing. (2010-05-12, Visual C++, 19513KB, 下载195次)
[模式识别(视觉/语音等)] AVehicleContourbasedMethodforOcclusion
摘要：在交通场景下进行多目标跟踪时，如何正确检测出车辆间的相互遮挡是影响车辆跟踪结果的关键。针对问题，运用投
影理论分析交通场景的三维几何投影特征．用长方体投影轮廓模型对车辆进行建模，重构其乏维投影轮廓，以进行遮挡的检
测和分离。与以往的方法相比，它在估计出的车辆外形轮廓基础t-进行遮挡检测，不需要匹配操作，计算量较小，并能解决
基于匹配的方法无法对付的初始遮挡问题。用实验验证了该算法的有效性。
In multi—object tracking of traf氍c scene。how to detect the occlusion is a
key problem for vehicle
tracking．A novel vehicle contour based method is prope∞d to deal with this problem．This method firstly extracts the
3D geometry character from the vehicle images according to the projecting theory，establishes the model of the vehicle
with the euboid project contour，and then reconstructs its 3D project contour。detects and segments the occlusion l-e-
sious．Compared with traditional method，it deals with the problem with less computation because it generates the ve·
hide contour firstly and no match calculation is needed．Moreover，this method is able to solve the initial occlusion
problem，which could not be solved with matching based method．The experimental results show that the proposed
method iS more emcient． (2009-09-18, Visual C++, 290KB, 下载128次)
[模式识别(视觉/语音等)] FisherFace
人脸识别的经典算法的完美结合，PAC与FISHER算法C++实现，首先通过PCA进行维数约简，然后通过FISHER进行最有利的方向投影。识别效率是所有监督学习的上限。
Face Recognition Algorithm for the perfect combination of classic, PAC and FISHER algorithm C++ Realize, first of all carried out through the PCA dimension reduction, and then through to the most favorable FISHER projection direction. Recognition efficiency is all supervised learning ceiling. (2008-09-27, Visual C++, 741KB, 下载216次)
[模式识别(视觉/语音等)] tiaoxingma
barcode: 条形码识别，直接运行程序即可；
pdf417lib：二维条形码打印（输出为ps格式的文件），在书中第6章二维条形码打印部分有程序使用的说明；
条形码生成器源程序：生成一维条形码，直接运行程序即可；
barcode : barcode recognition, can run the program directly; Pdf417lib : two-dimensional bar code printing (ps output files), in Chapter 6 of the book 2D bar code printing some of the procedures used; Barcode Generator source : one-dimensional bar code generation, direct operation procedures can be; (2006-03-27, Visual C++, 341KB, 下载131次)
[模式识别(视觉/语音等)] license plate recoginition
汽车牌照的自动定位和识别程序源代码
source code for automatic license plate location and recognition (2004-11-18, Visual C++, 2679KB, 下载3149次)