Marinai S., Fujisawa H. (eds.) Machine Learning in Document Analysis and Recognition
Подождите немного. Документ загружается.
428 A. Schlapbach and H. Bunke
29. Melin, H., Koolwaaij, J., Lindberg, J., Bimbot, F.: A comparative evaluation of
variance flooring techniques in HMM-based speaker verification. In: Proc. 5th
Int. Conf. on Spoken Language Processing. (1998) 2379–2382
30. Collobert, R., Bengio, S., Mari´ethoz, J.: Torch: a modular machine learning
software library. IDIAP-RR 46, IDIAP (2002)
31. Bernard, T.M., Manzanera, A.: Improved low complexity fully parallel thinning
algorithm. In: Proc. 10th Int. Conf. on Image Analysis and Processing. (1999)
215–220
32. Marti, U.V., Bunke, H.: Using a statistical language model to improve the per-
formance of an HMM-based cursive handwriting recognition system. Int. Journal
of Pattern Recognition and Artificial Intelligence 15 (2001) 65–90
33. Schlapbach, A., Bunke, H.: A writer identification and verification system using
HMM based recognizers. Pattern Analysis and Applications 10(1) (2007) 33–43
34. Marukatat, S., Arti`eres, T., Gallinari, P., Dorizzi, B.: Rejection measures for
handwriting sentence recognition. In: Proc. 8th Int. Conf. on Frontiers in Hand-
writing Recognition. (2002) 25–29
35. Pitrelli, J.F., Perrone, M.P.: Confidence modeling for verification post-
processing for handwriting recognition. In: Proc. 8th Int. Workshop on Frontiers
in Handwriting Recognition. (2002) 30–35
36. Pitrelli, J.F., Perrone, M.P.: Confidence-scoring post-processing for off-line
handwritten-character recognition verification. In: Proc. 7th Int. Conf. on Doc-
ument Analysis and Recognition. (2003) 278–282
37. Rosenberg, A.E., Deong, J., Lee, C.H., Juang, B.H., Soong, F.K.: The use of
cohort normalized scores for speaker verification. In: Proc. Int. Conf. on Spoken
Language Processing. (1992) 599–602
38. Matsui, T., Furui, S.: Likelihood normalization for speaker verification using a
phoneme- and speaker-independent model. Speech Communications 17 (1995)
109–116
39. Marti, U.V., Bunke, H.: The IAM–database: An English sentence database for
off-line handwriting recognition. Int. Journal of Document Analysis and Recog-
nition 5 (2002) 39–46
40. Kuncheva, L.I.: Combining pattern classifiers: methods and algorithms. Wiley-
Interscience (2004)
41. Bimbot, F., Chollet, G.: Assessement of speaker verification systems. In Gibbon,
D., Moore, R., Winski, R., eds.: Handbook of Standards and Resources for
Spoken Language Systems, Mouton de Gruyter (1997) 408–480
42. Plamondon, R., Srihari, S.N.: On-line and off-line handwriting recognition: A
comprehensive survey. IEEE Trans. on Pattern Analysis and Machine Intelli-
gence 22 (2000) 63–84
43. Barras, C., Gauvain, J.L.: Feature and score normalization for speaker verifica-
tion of cellular data. In: Int. Conf. on Acoustics, Speech, and Signal Processing.
Volume 2. (2003) 49–52
Index
active learning, 243
adaptive classifier, 233
address-block extraction, 281
ANN, see artificial neural network
artificial neural network, 23, 139, 142,
151, 194, 260, 261
ascender, 31, 203, 232, 243, 263, 345,
415
back-propagation, 269
back-propagation through structure, 26
bad segmentation, 269
bagging, 163, 372
baseline, 262, 340, 415
Bayes rule, 141, 288, 297
behavior knowledge space, 378
best impostor score distribution, 322
binarization, 242
biometric matchers, 308
boosting, 163, 373
bounded horizon, 226
capacity of recognition system, 335,
349, 353
CAPTCHA, 237, 337
CAVIAR, 241
cell alignment, 392
chain identification, 61
character recognition, 8, 270
Chinese, 145
hand-printed, 222
off-line, 177
on-line, 176
character segmentation, 11, 224, 259,
269, 273, 281, 293
character validation, 261
class label, 225
class-and-style conditional feature
distributions, 225
classification
document, 125
field, 226
font, 232
classifier
adaptive, 233
ensembles, 370
k-nn, see nearest neighbor classifier
MLP, see multi-layer perceptron
quadratic, 150
RBF, see radial basis function
statistical, 150
structure adaptive, 201
style-first, 232
SVM, see support vector machine
classifier combination, 74, 154, 163, 361
Borda count, 374
combination schemes, 374
complexity types, 366
majority voting, 374
output types, 365
schemes, 165
voting, 374
client, 420
clustering, 37, 146, 198, 235
character, 200, 204
fuzzy, 235
430 Index
k-means, 202, 235, 414
layout, 212
word, 208
cohort model, 418
combination complexity types, 324
combination function, 313
likelihood ratio, 310, 315
optimal
approximation of, 321
as a sum of logistic functions, 323
for identification systems, 314, 319
for verification systems, 311, 319
weighted sum, 315
combination rule, 313
common-source patterns, 221
completeness property, 52, 54
confidence
calculation, 265
fusion, 266
measure, 417
value, 287
connected component, 200, 236, 261,
283, 285, 295, 340–344
consistency property, 52, 54
context, 226
contour analysis, 271
correlation, 390
covariance matrix, 414
cross validation, 419
cursive script, 259
database
CEDAR, 269
CENPARMI, 148
ETL9B, 149
IAM, 16, 346, 349, 419
MARG, 15
MediaTeam, 15
MNIST, 14, 148
NIST, 14, 148, 225
UW, 14, 246
decision-based algorithm
comparison, 92
specification, 73, 77, 96
Dempster-Shafer theory of evidence,
377
density feature, 269
descender, 31, 203, 232, 243, 263, 345,
415
digit recognition, 201
digital library, 106, 118, 213, 246
discrete-style classifier, 232
distortion strength, 350
document
classification, 125
degradation, 336
image analysis, 221
image understanding, 46
retrieval, 210
document layout analysis, 45
durable interaction, 237
dynamic field classifier, 233
EM, see expectation-maximization
enhanced heuristic segmenter, 261
enhanced segmentation technique, 262
enhancement, 200
ephemeral interaction, 237
equal error rate, 423
expectation-maximization, 413
exploratory data analysis, 237
false acceptance rate, 423
false rejection rate, 423
feature extraction, 223, 416
feature matching, 139
feature selection, 153
field, 224
classification, 226
length, 224
field-trained classifiers, 231
first-rank-correct rate, 309
font classification, 232
forensic document analysis, 387
forgery, 388
skilled, 421
unskilled, 420
Gabor filter, 182
gamma distribution, 393
Gaussian mixture, 229
Gaussian mixture model, 184, 413
general learning, 388, 393
geometrical transformations, 340
glyph, 222
GMM, see gaussian mixture model
ground truth, 75
handwriting, 200
Index 431
handwriting generation, 334, 343
handwriting recognition, 176, 344
off-line, 333
heuristic rules, 262, 272
heuristic segmenter, 261, 265
hidden Markov model, 201, 344, 350
historical precision, 91, 98
historical recall, 91, 98
human-initiated, 237
human-machine interaction, 223
identification model, 325
and cohort normalization, 327
and rank based combinations, 326
and score normalization, 325, 327
and score set statistics, 325
and T-normalization, 327
for likelihood ratio combination, 329
for weighted sum combination, 328
identification systems, 309, 312
likelihood ratio combination for, 314
ILP, see inductive logic programming
image registration, 391
impostor, 420
incorrect segmentation points, 259, 270
incremental learning, 107, 155
independence assumption for scores in
identification trial, 316
inductive logic programming, 53, 119
information theory, 168, 377
informational confidence, 167
inter-class style, 228
inter-pattern class-feature dependence,
230
interoperability, 246
intra-class style, 228
invariance
by feature extraction, 195
by structure, 195
by training, 195
Japanese character recognition, 176
kernel methods, 112, 144, 312
KL, see Kullback-Leibler
knowledge-based technique, 261
Kolmogorov-Smirnov, 396
KS, see Kolmogorov-Smirnov
Kullback-Leibler, 397
landmark mapping, 391
language models, 227, 345, 365
large character set, 145, 155, 176
latent semantic indexing, 127
layout analysis, 6, 22, 74, 106, 199, 278
layout structure, 45
learning
active, 243
unsupervised, 222
learning vector quantization, 143, 195
learning without a teacher, 234
lexical models, 227
ligature detection, 261, 262
likelihood ratio, 296, 396
likelihood ratio combination
for identification systems, 314
for verification systems, 310
with best impostor density, 322
logical labeling, 46
logical structure, 7, 45, 74
LSI, see latent semantic indexing
LVQ, see learning vector quantization
machine-initiated, 237
mail address
reading system, 283
recognition, 280
maximum likelihood principle, 414
MDF, see modified direction feature
meta-data, 118
missed segmentation, 269
mixture decomposition, 234
MLP, see multi-layer perceptron
mobile text entry, 239
modified direction feature, 262, 267,
272, 273
modified quadratic discriminant
function, 141, 146
modified vertical histogram, 262
morphological models, 227
MQDF, see modified quadratic
discriminant function
multi-layer perceptron, 23, 142, 149,
194, 265, 269, 272, 273
multiple classifier systems, 145
multiple hypotheses, 280, 285
multistrategy learning, 106
432 Index
nearest neighbor classifier, 141, 177,
180, 182, 356, 363, 381
neural assistant, 261, 262, 265, 272
neural validation, 274
neural-based segmentation, 262
noisy characters, 271, 337
normalized Spearman footrule distance,
63
OCR, see optical character recognition
optical character recognition, 37, 139,
180, 203, 222, 277, 336
order-dependent inter-pattern depen-
dence, 230
outlier, 151, 376
over-segmentation, 140, 259, 261, 269
overfitting, 59, 196, 349, 414
overlapping characters, 271
page segmentation, 243
PCA, see Principal Component Analysis
PDF, see portable document format
performance
estimate, 175
function, 169
theorem, 173
perturbation model, 338
physical structure, 6, 74
portable document acquisition, 244
portable document format, 111
postscript format, 111
pre-processing, 5, 108, 145, 147, 156,
177, 199, 283
precision, 63, 65, 91, 205, 232, 246, 310
principal component analysis, 142, 149,
208
projection profile, 82, 263, 411
PS, see postscript format
quadratic classifier, 150
quadratic discriminant function, 141
radial basis function, 25, 142, 143, 149,
194
RBF, see radial basis function
reading order, 10, 23, 46
reading order induction problem, 52
recall, 63, 65, 91, 205, 232, 246, 310
receiver operator characteristic, 309,
400, 423
recognition rate, 38, 148, 150, 152, 156,
164, 167, 175, 176, 178, 182, 185,
214, 260, 261, 270, 272, 307, 334,
338, 345, 347, 350, 353, 377, 380,
412, 419
recognition strategy language, 77
region of interest, 279
reject neuron, 267
reject rate, 399
rejection, 91, 96, 151, 153, 273, 403,
417, 422
retraining, 379
ROC, see receiver operator characteris-
tic
RSL, see recognition strategy language
saturation of recognition system, 353
scaling, 196, 341, 415
score normalization, 375
informational confidence, 377
median absolute deviation, 376
min-max normalization, 376
partial recognition rates, 377
z-score, 376
script identification, 12, 182
segmentation, 224
segmentation areas, 265
segmentation path detection, 259, 261,
268
segmentation performance, 269
segmentation point validation, 259, 261
self-organization, 194, 234
self-organizing map, 23, 194, 196
semantic web, 106
shearing, 340
signature representation, 389
signature verification, 12
simulated data, 225
skew detection, 243, 339
skilled forgery, 421
sliding window, 416
SOM, see self-organizing map
source label, 225
special learning, 388
specific learning, 393
statistical classifier, 150, 231
statistical dependence, 221
statistical methods, 141
structure adaptive classifiers, 201
Index 433
style, 228
style label, 225
style-first classifier, 232
support vector machine, 144, 149, 151,
184, 312
SVM, see support vector machine
syntactic models, 227
synthetic training data, 333
synthetically generated text, 335
system
ATRE, 53
CEDAR FOX, 405
DOMINUS, 136
WISDOM++ , 50
table recognition, 73, 75
tangent distance, 196, 204
template matching, 139
test set, 148, 156, 222, 224, 260, 270,
271, 307, 316, 346–349, 419, 420,
422
text line normalization, 415
tf-idf, 128, 209
thickening, 342
thinning, 199, 342, 416
timedelayneuralnetwork,9,25
topological sorting, 49
training of recognizers, 336
training set, 45, 132, 168, 174, 180, 203,
206, 214, 216, 222, 224, 270, 307,
310, 312, 324, 346, 347, 364, 366,
370–373, 378
expansion, 334, 336, 347, 349
generation, 153
variability, 334
transition feature, 265
underlying functions, 339
unskilled forgery, 420
unsupervised learning, 30, 143, 193,
194, 196, 198, 222, 234
validation set, 174, 222, 349, 350, 352,
353, 374, 401, 421
variance flooring, 414
vector quantization, 143, 235, 414
vectorization, 240
verification systems, 309, 310
likelihood ratio combination for, 310
virtual keyboard, 239
VQ, see vector quantization
Wang notation, 247
Web-wide data accessibility, 244
weighted Euclidean distance, 183
weighted sum combination rule, 315
word
-completion, 240
clustering, 208
indexing, 203
recognition, 9, 31, 32, 231, 260, 261,
306, 313, 362
segmentation, 243
shape feature, 390
world model, 418
writer identification, 12, 409
off-line, 409
on-line, 409
text dependent, 409
text independent, 409
writer verification, 409
off-line, 409
on-line, 409
text dependent, 409
text independent, 409
writer-independent, 346
XY tree, 48, 212