top-hat

基于模板匹配的另一种人脸识别的算法，很实用，利用肤色建模来建立人脸皮肤模型(Another based on template matching algorithm for face recognition, very practical, using skin color modeling to create a human face skin model)

几种经典的图像噪声方差估计方法（论文和源代码）(Several classical noise estimation method (articles and source code))

根据D. Comaniciu, P. Meer: Mean Shift: A robust approach toward feature space analysis 以及 C. Christoudias, B. Georgescu, P. Meer: Synergism in low level vision.这两篇文献提供的方法编写的图像分割代码，作者是 Chris M. Christoudias, Bogdan Georgescu，代码经我看了后加了丰富的中文注释，希望可以给各位带来阅读上的方便。 基于meanshift的聚类分割方法包括滤波、区域融合等操作，通过调整sigma和sigmar来调整分割效果。(According to D. Comaniciu, P. Meer: Mean Shift: A robust approach toward feature space analysis, and C. Christoudias, B. Georgescu, P. Meer: Synergism in low level vision. These two documents prepared by the methods provided by image segmentation code , the author is Chris M. Christoudias, Bogdan Georgescu, after I read the code, add a rich Chinese notes, hoping to bring you the convenience of reading. Segmentation method based on clustering meanshift including filtering, regional integration and other operations, and by adjusting the sigma sigmar to adjust segmentation results.)

稀疏矩阵字典的优化KSVD，本人已进行测试，效果甚佳,可以用在图像，数据的压缩和分类(Sparse dictionary optimization KSVD, I have been tested, very good results)

将2D图像转换为红蓝3D效果。程序由本人编写，算法来自论文《A novel 2D-to-3D conversion technique based on relative height depth cue》by Yong Ju Jung等。(2D-to-3D Conversion. The codes are written by myself and according to the article "A novel 2D-to-3D conversion technique based on relative height depth cue" by Yong Ju Jung.)

对立体视觉的双目图像进行极线校正，以实现双目致密匹配(Of three-dimensional visual image of the binocular epipolar rectification to realize dense binocular matching)

用与水蒸气冷凝为水的程序，可以作为源项加入质量方程和能量方程(condensation UDF)

1、参数初始化； 2、形成边界削波数据； 3、波场初始化； 4、Zmax层波场延拓到深度Zmax-1； 5、Zi+1层波场延拓到深度Zi； 6、重复5，从Iz Nz-1开始，直到Iz 1，得测线上的频率—空间域波场； 7、频率-空间域波场对频率做反傅里叶变换，得时间-空间波场； (1, parameter initialization 2, the formation of boundary clipping data 3, wave field initialization 4, Zmax layer wave field extended to the depth of Zmax-1 5, Zi+1 layer wave field extended to the depth of Zi 6, repeat 5, starting Iz Nz-1 until Iz 1, the frequency of the test line- spatial domain wave field 7, frequency space domain wave field to frequency inverse Fourier transform, time- space field )

梯度法计算指纹方向场，采用了三种方式：1、滑动块 2、自适应块 3、分割固定块(Gradient method is used to calculate the direction field of fingerprint, and three methods are adopted: 1, sliding block 2, adaptive block 3 and segmentation fixed block)

基于图像检索的matlab代码，基于内容的图像检索中的一些关键环节：特征提取：颜色直方图；纹理特征等 相似度：马氏距离，欧氏距离等 相关反馈：机器学习方法，如SVM，神经网络等 检索与分类：两个很相似的样本距离很小，虽然两个不相似的样本距离未必很大(content-based image retrieval of some of the key issues : Feature Extraction : color histogram Texture characteristics of similarity : Mahalanobis distance, the Euclidean distance relevance feedback : machine learning methods, such as SVM. Neural network search and classification : two very similar samples from the small, although the two are similar to the samples may not be much distance)