fingerprint-classification3

卡夫曼算法，目标跟踪，可预测。简单算法，我在后面增加了汉语注释，方便初学者了解。(Target tracking using kalman，using this matlab program, you can see a ball tracked with a lable.)

基于马尔科夫链的图像分割，较以往经典图像分割性能有很大提高(Markov chain-based image segmentation, image segmentation performance than the classical past has greatly improved)

基于区域的空间域图像融合。先对源图像作小波分解，低频分量加权平均，高频分量用模糊C均值聚类算法进行区域分割，对区域进行基于ssim值的融合，最后小波逆变换得到融合图像。(region based spatial domain method.First do the source image wavelet decomposition low frequency components weighted average of the high frequency component using Fuzzy C-means clustering algorithm region segmentation of regional integration based ssim value, and finally fused image inverse wavelet transform)

说明：  Arnold 变换与 Gray 码变换相融合的双置乱算法程序，包括了加密和解密程序(The program of double scrambling algorithm combining Arnold transform and gray code transform, including encryption and decryption program)

基于互信息的图像配准源代码，，采用粒子群优化算法可直接使用。 (Based on mutual information image registration source code, using particle swarm optimization algorithm can be used directly.)

说明：  应用三维重构原理对二维平面图形重建成三维图像。(Application of the principle of three-dimensional reconstruction of two-dimensional graphical reconstruction into three-dimensional image.)

常用的图像去噪算法，基于隐马尔可夫树模型的小波去噪 基于尺度空间和context模型相结合的自适应小波去噪 基于尺度和空间混合模型的小波图像去噪 Bayes估计阈值软门限去噪 (The denoising algorithm used in image denoising, wavelet hidden Markov tree model based on adaptive wavelet denoising combining the scale space and context model based on wavelet denoising image scale and spatial mixed model based on Bayes estimation soft threshold denoising )

1、对观测数据进行中心化，； 2、使它的均值为0，对数据进行白化—>Z； 3、选择需要估计的分量的个数m，设置迭代次数p<-1 4、选择一个初始权矢量（随机的W，使其维数为Z的行向量个数）； 5、利用迭代W(i,p)=mean(z(i,:).*(tanh((temp) *z)))-(mean(1-(tanh((temp)) *z).^2)).*temp(i,1)来学习W （这个公式是用来逼近负熵的） 6、用对称正交法处理下W 7、归一化W(:,p)=W(:,p)/norm(W(:,p)) 8、若W不收敛，返回第5步 9、令p=p+1，若p小于等于m,返回第4步 剩下的应该都能看懂了 基本就是基于负熵最大的快速独立分量分析算法(1, on the center of the observation data, 2, making a mean of 0, the data to whitening-> Z 3, select the number of components to be estimated m, setting the number of iterations p < -1 4, select an initial weight vector (random W, so that the Z dimension of the row vectors of numbers) 5, the use of iteration W (i, p) = mean (z (i, :).* (tanh ((temp) ' * z)))- (mean (1- (tanh ((temp)) ' * z). ^ 2)).* temp (i, 1) to learn W (This formula is used to approximate the negative entropy) 6 with symmetric orthogonal treatments W 7, normalized W (:, p) = W (:, p)/norm (W (:, p)) 8, if W does not converge, return to step 5 9 , so that p = p+1, if p less than or equal m, return to step 4 should be able to read the rest of the basic is based on negative entropy of the largest fast independent component analysis algorithm)

matlab关于DCT的图像处理程序 供大家交流(Matlab on DCT image processing procedures for the exchange of)

用于单目标追踪的多模概率假设粒子滤波程序(MM-PHD for single target)