Description: 现有的代数特征的抽取方法绝大多数采用一维的方法，即首先将图像转换为一维向量，再用主分量分析（PCA）,Fisher线性鉴别分析（LDA）,Fisherfaces式核主分量分析（KPCA）等方法抽取特征，然后用适合的分类器分类。针对一维方法维数过高，计算量大，协方差矩阵常常是奇异矩阵等不足，提出了二维的图像特征抽取方法，计算量小，协方差矩阵一般是可逆的，且识别率较高。-existing algebra feature extraction method using a majority of the peacekeepers, First images will be converted into one-dimensional vector, and then principal component analysis (PCA), Fisher Linear Discriminant Analysis (LDA), Fisherfaces audits principal component analysis (KPCA), and other selected characteristics, then use the appropriate classification for classification. Victoria against an excessive dimension method, calculation, covariance matrix is often inadequate singular matrix, a two-dimensional image feature extraction method, a small amount of covariance matrix is usually reversible, and the recognition rate higher. Platform: |
Size: 2513 |
Author:小弟 |
Hits:
Description: 现有的代数特征的抽取方法绝大多数采用一维的方法，即首先将图像转换为一维向量，再用主分量分析（PCA）,Fisher线性鉴别分析（LDA）,Fisherfaces式核主分量分析（KPCA）等方法抽取特征，然后用适合的分类器分类。针对一维方法维数过高，计算量大，协方差矩阵常常是奇异矩阵等不足，提出了二维的图像特征抽取方法，计算量小，协方差矩阵一般是可逆的，且识别率较高。-existing algebra feature extraction method using a majority of the peacekeepers, First images will be converted into one-dimensional vector, and then principal component analysis (PCA), Fisher Linear Discriminant Analysis (LDA), Fisherfaces audits principal component analysis (KPCA), and other selected characteristics, then use the appropriate classification for classification. Victoria against an excessive dimension method, calculation, covariance matrix is often inadequate singular matrix, a two-dimensional image feature extraction method, a small amount of covariance matrix is usually reversible, and the recognition rate higher. Platform: |
Size: 2048 |
Author:小弟 |
Hits:
Description: （1）应用9×9的窗口对上述图象进行随机抽样，共抽样200块子图象；
（2）将所有子图象按列相接变成一个81维的行向量；
（3）对所有200个行向量进行KL变换，求出其对应的协方差矩阵的特征向量和特征值，按降序排列特征值以及所对应的特征向量；
（4）选择前40个最大特征值所对应的特征向量作为主元，将原图象块向这40个特征向量上投影，所获得的投影系数就是这个子块的特征向量。
（5）求出所有子块的特征向量。
-(1) the application of 9 × 9 window of these images at random, a total sample of 200 sub-image (2) all sub-images according to out-phase into a 81-dimensional row vector (3) all 200 lines for KL transform vector, derived its corresponding covariance matrix of eigenvectors and eigenvalues, in descending order by eigenvalue and the corresponding eigenvector (4) a choice to 40 corresponding to the largest eigenvalue eigenvector as the PCA, the original image block to the 40 feature vectors on the projection, the projection coefficients obtained by this sub-block eigenvector. (5) calculated for all sub-block eigenvector. Platform: |
Size: 64512 |
Author:ly |
Hits:
Description: 极化SAR图像处理中，将协方差矩阵转换为STOKES矩阵或Mueller矩阵的程序-Polarization SAR image processing, the covariance matrix will be converted to matrix or Mueller matrix STOKES procedures Platform: |
Size: 1024 |
Author:王文光 |
Hits:
Description: （1）应用9×9的窗口对上述图象进行随机抽样，共抽样200块子图象；
（2）将所有子图象按列相接变成一个81维的行向量；
（3）对所有200个行向量进行KL变换，求出其对应的协方差矩阵的特征向量和特征值，按降序排列特征值以及所对应的特征向量；
（4）选择前40个最大特征值所对应的特征向量作为主元，将原图象块向这40个特征向量上投影，所获得的投影系数就是这个子块的特征向量。
（5）求出所有子块的特征向量。
-(1) the application of 9 × 9 window of these images at random, a total sample of 200 sub-image (2) all sub-image by out-phase into a 81-dimensional vector lines (3) All 200 line vector KL transform, derive its corresponding covariance matrix eigenvectors and eigenvalues, in descending order eigenvalues and corresponding eigenvectors (4) a choice to 40 corresponding to the largest eigenvalue eigenvector as the main element, the original image block to the 40 on the projection eigenvector obtained projection coefficient is the sub-block eigenvector. (5) calculated for all sub-block eigenvector. Platform: |
Size: 2048 |
Author:龙活 |
Hits:
Description: 计算SENSE重建图像中的g-factor,这是并行磁共振成像SENSE算法的关键一步-G-factor is the metric to quantify the amplificaiton of noise power in reconstructing SENSE accelerated image. The detail was presented in Pruessmann s 1999 Magn. Reson. Med. paper. In theory, g-factor is the pixel-by-pixel ratio of the image variance between the SENSE reconstructed image and un-accelerated image. And g-factor depends on: (1) the k-space data trajectory (2) coil sensitivit profiles (3) noise covariance matrix of the array coil Platform: |
Size: 5120 |
Author:李荣智 |
Hits:
Description: 经典的最大似然法分类法的C语言实现，有助于深入了解遥感分类原理。-This program implements the maximum likelihood classification procedure. ouput:1.classified image, and 2. probability file.
Note: For constructong variance-covariance matrix must be generic binary file.
Platform: |
Size: 4096 |
Author:李会利 |
Hits:
Description: 1.对一个256*256的图像进行DCT变换得到图像D，将D得斜下角数值置为零，然后进行DCT反变换.
2.对源图像进行K-L转换
1和2比较-1.Get a grey level image which size is N*N. (For example, 256*256, however,
N = ), and partition to 8*8 sub images.
2.. Apply DCT to these sub images, and get the transformed image D with DCT
coefficients for elements.
3. From D, keep the coefficient values for only upper left triangular region and set
zeros for lower right region to approximate the image. (That is, only half of data
is used.)
4.Take Inverse DCT to get the approximated image.
2 . Get the covariance matrix of image.
3 . Calculate the corresponding eigenvectors and eigenvalues.
4 . Represent the original image with Singular Value Decomposition.
5 . Approximate the image by taking off the 4 smallest eigenvalues. (That is, only
half of information is used.)
Platform: |
Size: 1024 |
Author:zhengyan |
Hits:
Description: 为了更有效地提取图像的局部特征，提出了一种基于2维偏最小二乘法(two—dimensional partial least
square，2DPLS)的图像局部特征提取方法，并将其应用于面部表情识别中。该方法首先利用局部二元模式(1ocal
binary pattern，LBP)算子提取一幅图像中所有子块的纹理特征，并将其组合成局部纹理特征矩阵。由于样本图像
被转化为局部纹理特征矩阵，因此可将传统PLS方法推广为2DPLS方法，用来提取其中的判别信息。2DPLS方法
通过对类成员关系矩阵的构造进行相应的修改，使其适应样本的矩阵形式，并能体现出人脸局部信息重要性的差
异。同时，对于类成员关系协方差矩阵的奇异性问题，也推导出了其广义逆的解析解。基于JAFFE人脸表情库的
实验结果表明，该方法不但可以有效地提取图像局部特征，并能取得良好的表情识别效果。-To better the image of the local feature extraction, a partial least squares method based on 2D (two-dimensional partial least
square, 2DPLS) image local feature extraction method, and applied to facial expression recognition. In this method, use of local binary pattern (1ocal
binary pattern, LBP) operator extracts an image texture features of all sub-blocks, and their combination into the local texture feature matrix. As the sample image
Be translated into the local texture feature matrix, so the traditional PLS method can be generalized to 2DPLS method used to extract the identification information. 2DPLS method
Through the class membership matrix in the corresponding modifications to adapt the sample matrix, and can reflect the importance of face poor local information
Different. Meanwhile, members of the class covariance matrix of the singular relations issues, also derived the generalized inverse of the analytical solution. Based on the JAFFE facial expression database
Platform: |
Size: 315392 |
Author:MJ |
Hits:
Description: 在神经网络中用求协方差矩阵的方法对人脸图像进行压缩及恢复-In the neural network covariance matrix using the method of face image compression and restoration Platform: |
Size: 1024 |
Author:曾强 |
Hits:
Description: Decompose image into subbands (undecimated wavelet), denoise, and recompose again.
fh = decomp_reconst_wavelet(im,Nsc,daub_order,block,noise,parent,covariance,optim,sig)
im : image
Nsc: Number of scales
daub_order: Order of the daubechie fucntion used (must be even).
block: size of neighborhood within each undecimated subband.
noise: image having the same autocorrelation as the noise (e.g., a delta, for white noise)
parent: are we including the coefficient at the central location at the next coarser scale?
covariance: are we considering covariance or just variance?
optim: for choosing between BLS-GSM (optim = 1) and MAP-GSM (optim = 0)
sig: standard deviation (scalar for uniform noise or matrix for spatially varying noise)
Javier Portilla, Univ. de Granada, 3/03
Revised: 11/04 - Decompose image into subbands (undecimated wavelet), denoise, and recompose again.
fh = decomp_reconst_wavelet(im,Nsc,daub_order,block,noise,parent,covariance,optim,sig)
im : image
Nsc: Number of scales
daub_order: Order of the daubechie fucntion used (must be even).
block: size of neighborhood within each undecimated subband.
noise: image having the same autocorrelation as the noise (e.g., a delta, for white noise)
parent: are we including the coefficient at the central location at the next coarser scale?
covariance: are we considering covariance or just variance?
optim: for choosing between BLS-GSM (optim = 1) and MAP-GSM (optim = 0)
sig: standard deviation (scalar for uniform noise or matrix for spatially varying noise)
Javier Portilla, Univ. de Granada, 3/03
Revised: 11/04 Platform: |
Size: 1024 |
Author:ali |
Hits:
Description: This paper presents a new human skin color model in
YCbCr color space and its application to human face
detection. Skin colors are modeled by a set of three
Gaussian clusters, each of which is characterized by a
centroid and a covariance matrix. The centroids and
covariance matrices are estimated from large set of
training samples after a k-means clustering process. Pixels
in a color input image can be classified into skin or nonskin
based on the Mahalanobis distances to the three
clusters. Efficient post-processing techniques namely noise
removal, shape criteria, elliptic curve fitting and faceinonface
classification are proposed in order to !inther refine
skin segmentation results for the purpose of face detection. Platform: |
Size: 220160 |
Author:陆 |
Hits:
Description: 这是一个多重脸识别的程序，代码采用k_L变换和奇异值分解，利用协方差矩阵获得人脸图像的特征脸。-This is a multi-face recognition program, the code used k_L transform and singular value decomposition, the covariance matrix can be obtained by using the face image features the face. Platform: |
Size: 131072 |
Author:刘晖 |
Hits:
Description: 基于内容的图像检索技术的关键在于特征提取，是利用图像的颜色、形状、纹理、轮廓、对象的空间关系等客观独立的存在于图像中的基本视觉特征作为图像的索引，计算查询图像和目标图像的相似距离，按相似度匹配进行检索。综合国内外研究现状，可将基于内容的图像检索技术分为如下几种类型:基于颜色特征的检索、基于纹理特征的检索、基于形状及区域的检索、基于空间约束关系的检索。-Based on comparing various affine invariant regional basis, selection of image stabilization exremum area biggest content segmentation and extraction. It has the affine invariants, the neighboring territory, stability and multi-scale characteristics, but also because of regional only by grey value of decision, so the size relations is not sensitive illumination change. In images all the pixel, then through sorting for barrel separated binary tree forest-- set the extreme area all obtained images and construct component tree, finally through the biggest stable delay-independent conditions MSER area is MSER, obtain the area without rules boundary shape of its, in order to facilitate to quantification description, using covariance matrix region neat optimization, made the final output of extreme value for the oval areas regional. Platform: |
Size: 2584576 |
Author:陈利华 |
Hits:
Description: An imlementation in MATLAB of fast algorithm for calculate covariance matrix, which is widely used in image processingimlementation of fast algorithm for calculate covariance matrix, which is widely used in image processing Platform: |
Size: 1781760 |
Author:Co |
Hits:
Description: 经典的最大似然法分类法的C语言实现，有助于深入了解遥感分类原理。-This program implements the maximum likelihood classification procedure. ouput:1.classified image, and 2. probability file.Note: For constructong variance-covariance matrix must be generic binary file. Platform: |
Size: 3072 |
Author:whicme |
Hits:
Description: 研究表明，构成K一L变换矩阵的向量也就是ToePhtz矩阵的特征向量。与此同时，离散
余弦变换矩阵逼近于ToePutz矩阵的特征向量矩阵，所以离散余弦变换矩阵与自然图像的
K一L变换矩阵十分相似。经过离散余弦变换后的变换系数块的协方差矩阵Cy非常接近对角
阵，即除了对角线元素以外，其它很多元素都近似为0，并且在左上角集中了主要能量。这
反映了自然图像大部分区域变化不大，亮度突变只占少数，即图像能量以低频成分为主的特
性。通过变换后的量化，舍弃对视觉效果影响较小的次要信息，可达到进一步的压缩效果。
虽然从去相关性能的意义上讲，DCT是一种次于K一L变换的准最佳变换，但是从算法实现的角度来看，DCT则远远优于K一L变换。首先，当图像的分块大小确定后，DCT的变换矩阵也就随之确定了，不随输入信号的统计特性变化而变化 其次，二维DCT能够分解成两次一维DCT，有利于硬件实现。
-Studies have shown that constitute a K L transformation matrix vector is ToePhtz eigenvector of the matrix. At the same time, discrete
The cosine transform matrix approximation in ToePutz matrix eigenvector matrix, so the discrete cosine transform matrix and natural images
Is very similar to a K L transformation matrix. After discrete cosine transform, the transform coefficient block of Cy is very close to the diagonal of covariance matrix
Matrix, namely except diagonal elements, many other elements are approximate to 0, and focused the main energy in the upper left corner. this
Reflects the natural images most area changed little, brightness mutations accounted for only a few, the energy is given priority to with low frequency components of image
Sex. Through quantitative after transformation, abandoning smaller effect on the visual performance of secondary information, can achieve further compression effect.
Although from related to the performance of the sense, DCT is a kind o Platform: |
Size: 546816 |
Author:chen |
Hits:
Description: 两个代码一个是demo, demo是它的小样例子, 另外一个是它的源代码.
- -
This is the matlab implementation of following noise level estimation:
Xinhao Liu, Masayuki Tanaka and Masatoshi Okutomi
Noise Level Estimation Using Weak Textured Patches of a Single Noisy Image
IEEE International Conference on Image Processing (ICIP), 2012.
-
Copyright (C) 2012 Masayuki Tanaka. All rights reserved.
mtanaka@ctrl.titech.ac.jp
-
Contents
-
* NoiseLevel.m
The main code of the noise level estimation.
You can show the description by
> help NoiseLevel
demo.m also includes simple usage.
This algorithm is implemented with only single m file.
* demo.m
Demonstration example.
* lena.png
Sample image.
* README.txt
This file.
-
Note
-
We used the maximum eigenvalue of the gradient covariance matrix Platform: |
Size: 3072 |
Author:马欢 |
Hits:
Description: PCA的思想为将图像的协方差矩阵分解，获得分解后的方向向量。然后将数据分别投影到某一个方向上去，获得与原图象近似的图像。当然,与最大特征值所对应的特征向量方向获得最好的图像。因此，PCA方法可以作为降维的一种方法。留下在某些方向较好的图像，而抛弃那些在另外一些方向上不好的图像。-PCA ideas as to decompose the covariance matrix of the image, the direction vector obtained after decomposition. The data is then separately projected up to a certain direction, to obtain an image similar to the original image. Of course, with a maximum value corresponding to the characteristic feature vector direction to get the best image. Therefore, PCA method can be used as a method of dimensionality reduction. Leave a better image in some directions, and discard those in the other direction a bad image. Platform: |
Size: 1024 |
Author:fuliting |
Hits:
Search - Image covariance Matrix