PR_AI_pic
所属分类：图形图像处理
开发工具：Visual C++
文件大小：24331KB
下载次数：72
上传日期：2009-02-28 16:51:14
上 传 者：
局长
说明： 这个资源是《精通VC++数字图像模式识别技术及工程实践（第2版）》光盘中的实验素材，包括人脸图像、指纹图像、数字和字母图像、USPS图像。
(The resources, " proficient in VC++ Digital Image Pattern Recognition Technology and Engineering Practice [2nd edition]" CD-ROM in the experimental material, including the face image, fingerprint images, images of numbers and letters, USPS images.)
文件列表:
实验素材\USPS字库\save_load_matrix (40857, 2000-05-16)
实验素材\USPS字库\save_load_matrix_little.C (26745, 2000-05-16)
实验素材\USPS字库\test.dat (513792, 2001-06-12)
实验素材\USPS字库\testtarg.dat (2007, 2001-06-12)
实验素材\USPS字库\train.dat (1866496, 2001-06-12)
实验素材\USPS字库\traintarg.dat (7291, 2001-06-12)
实验素材\USPS字库\trans.m (211, 2001-06-12)
实验素材\USPS字库\trans1.m (257, 2001-06-12)
实验素材\USPS字库\tt.m (122, 2001-06-11)
实验素材\USPS字库\USPSfulldatabase.mat (19792370, 2000-05-17)
实验素材\USPS字库\USPStestingdata.mat (4277123, 2000-05-17)
实验素材\USPS字库\USPStrainingdata.mat (15521477, 2000-05-17)
实验素材\人脸图像\100x100\s1\1.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s1\10.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s1\2.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s1\3.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s1\4.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s1\5.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s1\6.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s1\7.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s1\8.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s1\9.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s1\Thumbs.db (46080, 2002-07-18)
实验素材\人脸图像\100x100\s10\1.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s10\10.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s10\2.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s10\3.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s10\4.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s10\5.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s10\6.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s10\7.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s10\8.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s10\9.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s10\Thumbs.db (36352, 2002-07-18)
实验素材\人脸图像\100x100\s11\1.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s11\10.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s11\2.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s11\3.bmp (11078, 2002-03-29)
实验素材\人脸图像\100x100\s11\4.bmp (11078, 2002-03-29)
... ...
US Postal Service Database
==========================
This database contains 7291 training patterns and 2007 test
patterns, with (16x16) resolution. All patterns are
0-9 digit representations.
USPS data were saved using [-1,+1] range to represent patterns
and the target is given by a number +1 in the correct position
of the target vector and -1 to the other classes.
The original files can be found on
http://www.kernel-machines.org/
The original binaries are in big endian format and must be
converted to run on intel platforms. No Matlab binaries
are provided.
The original files were converted to little endian and
saved in Matlab binary format (version 4). Matlab checks
binary files when reading to detect big or little endian.
TRAINING
========
The training data are organised by rows using a (7291x256)
matrix saved in Matlab binary format. Targets are stored
in a (7291x10) matrix.
------------------------------------------------
Name Size Bytes Class
------------------------------------------------
cmap 256x3 6144 double array
traindata 7291x256 14931968 double array
traintarg 7291x10 583280 double array
------------------------------------------------
Table 1: Variables in file USPStrainingdata.mat
USPS data were saved using [-1,+1] range to represent patterns.
If you want to use grey levels is interesting to represent them
using [0,255] range. This can be done like this:
% first, load all patterns
load USPStrainingdata;
% normalise the traindata in the [0,255] range
traindata = 255*(traindata+1)/2;
% assign [1 1 1] to white and [0 0 0] to black
traindata = round(255-traindata);
% plot a pattern (eg, the first pattern, six number)
i = 1;
image(reshape(traindata(i,:),16,16)');
colormap(cmap);
axis off;
The variable traintarg gives which digit each pattern represents.
This is done by assigning 1 to the correct class and -1 to
the other classes. For instance, for the previous pattern,
traintarg(1,:) will provide:
targets 0 1 2 3 4 5 6 7 8 9
traintarg(1,:) => [-1 -1 -1 -1 -1 -1 1 -1 -1 -1]
^
|
+--- number six
The colormap was created with the following commands:
% create a colormap for 256 Gary levels:
cmap = [0:255; 0:255; 0:255]';
cmap = cmap./255;
You can save memory using only uint8 to represent each pixel
instead double (eight times more memory is required) and a
single (7291x1) uint8 vector to represent the target class
(putting the digits 0-9 as targets)
TESTING
=======
The testing data are organised by rows using a (2007x256)
matrix saved in Matlab binary format. Targets are stored
in a (2007x10) matrix.
------------------------------------------------
Name Size Bytes Class
------------------------------------------------
cmap 256x3 6144 double array
testdata 2007x256 4110336 double array
testtarg 2007x10 160560 double array
------------------------------------------------
Table 2: Variables in file USPStestingdata.mat
All comments in previous section can be applied
to this section as well.
FULL DATABASE
=============
The file USPSfulldatabase.mat contains the full database
without distinction between training and testing data.
------------------------------------------------
Name Size Bytes Class
------------------------------------------------
cmap 256x3 6144 double array
patterns 92***x256 19042304 double array
targets 92***x10 743840 double array
------------------------------------------------
Table 3: Variables in file USPSfulldatabase.mat
COMMENTS
========
Marcelo Barros de Almeida
barros@cpdee.ufmg.br
16/May/2000
近期下载者：
相关文件：
收藏者：