32:
of two objects on the basis of sets of measurements for each object and a statistical model. In a classification procedure, the class for a new object (whose real class is unknown) can be estimated by minimising, across classes, an average of the Fisher kernel distance from the new object to each
374:. The FV encoding stores the mean and the covariance deviation vectors per component k of the Gaussian-Mixture-Model (GMM) and each element of the local feature descriptors together. In a systematic comparison, FV outperformed all compared encoding methods (
378:, Kernel Codebook encoding (KCB), Locality Constrained Linear Coding (LLC), Vector of Locally Aggregated Descriptors (VLAD)) showing that the encoding of second order information (aka codeword covariances) indeed benefits classification performance.
293:
341:
The Fisher kernel is the kernel for a generative probabilistic model. As such, it constitutes a bridge between generative and probabilistic models of documents. Fisher kernels exist for numerous models, notably
366:
representation suffers from sparsity and high dimensionality. The Fisher kernel can result in a compact and dense representation, which is more desirable for image classification and retrieval problems.
147:
321:
616:
370:
The Fisher Vector (FV), a special, approximate, and improved case of the general Fisher kernel, is an image representation obtained by pooling local image
351:
190:
417:
375:
363:
453:
595:
362:
The Fisher kernel can also be applied to image representation for classification or retrieval problems. Currently, the most popular
86:
496:
A.P. Twinanda et al. (2014), “Fisher Kernel Based Task
Boundary Retrieval in Laparoscopic Database with Single Video Query”
45:
17:
387:
371:
478:
Florent
Perronnin and Christopher Dance (2007), “Fisher Kernels on Visual Vocabularies for Image Categorization”
600:
49:
532:"Plant species classification using flower images—A comparative study of local feature representations"
436:
302:
405:
41:
462:
325:
76:
56:
generative models can process data of variable length (adding or removing data is well-supported)
29:
591:
571:
553:
530:
Seeland, Marco; Rzanny, Michael; Alaqraa, Nedal; Wäldchen, Jana; Mäder, Patrick (2017-02-24).
487:
Herve Jegou et al. (2010), “Aggregating local descriptors into a compact image representation”
413:
561:
543:
37:
408:
and David
Haussler (1998), Exploiting Generative Models in Discriminative Classifiers. In
437:"Generative vs Discriminative Approaches to Entity Recognition from Label-Deficient Data"
424:
566:
531:
169:
420:
610:
506:
25:
440:
JADT 2004, 7èmes journées internationales analyse statistique des données textuelles
588:
An
Introduction to Support Vector Machines and other kernel-based learning methods
548:
347:
343:
557:
575:
59:
discriminative methods can have flexible criteria and yield better results.
288:{\displaystyle K(X_{i},X_{j})=U_{X_{i}}^{T}{\mathcal {I}}^{-1}U_{X_{j}}}
36:
The Fisher kernel was introduced in 1998. It combines the advantages of
435:
Cyril Goutte, Eric
Gaussier, Nicola Cancedda, Hervé Dejean (2004))
308:
254:
156:
being a set (vector) of parameters. The function taking
142:{\displaystyle U_{X}=\nabla _{\theta }\log P(X|\theta )}
305:
193:
89:
410:
315:
287:
141:
8:
442:, Louvain-la-Neuve, Belgium, 10-12 mars 2004
586:Nello Cristianini and John Shawe-Taylor.
565:
547:
307:
306:
304:
277:
272:
259:
253:
252:
245:
238:
233:
217:
204:
192:
128:
107:
94:
88:
398:
352:probabilistic latent semantic analysis
7:
590:. Cambridge University Press, 2000.
617:Kernel methods for machine learning
507:"VLFeat - Documentation > C API"
73:The Fisher kernel makes use of the
455:Deriving TF-IDF as a fisher kernel
412:, pages 487–493. MIT Press.
358:Image classification and retrieval
104:
14:
33:known member of the given class.
316:{\displaystyle {\mathcal {I}}}
223:
197:
136:
129:
122:
1:
38:generative statistical models
549:10.1371/journal.pone.0170629
172:of the probabilistic model.
633:
18:statistical classification
388:Fisher information metric
376:Bag of Visual Words (BoW)
461:. SPIRE. Archived from
50:support vector machines
30:measures the similarity
452:Charles Elkan (2005).
317:
289:
143:
46:discriminative methods
468:on December 20, 2013.
337:Information retrieval
318:
290:
144:
28:, is a function that
394:Notes and references
303:
191:
87:
364:bag-of-visual-words
250:
42:hidden Markov model
326:Fisher information
313:
285:
229:
139:
418:978-0-262-11245-1
624:
580:
579:
569:
551:
527:
521:
520:
518:
517:
503:
497:
494:
488:
485:
479:
476:
470:
469:
467:
460:
449:
443:
433:
427:
403:
322:
320:
319:
314:
312:
311:
294:
292:
291:
286:
284:
283:
282:
281:
267:
266:
258:
257:
249:
244:
243:
242:
222:
221:
209:
208:
148:
146:
145:
140:
132:
112:
111:
99:
98:
632:
631:
627:
626:
625:
623:
622:
621:
607:
606:
583:
542:(2): e0170629.
529:
528:
524:
515:
513:
505:
504:
500:
495:
491:
486:
482:
477:
473:
465:
458:
451:
450:
446:
434:
430:
404:
400:
396:
384:
360:
339:
334:
301:
300:
273:
268:
251:
234:
213:
200:
189:
188:
178:
103:
90:
85:
84:
71:
66:
44:) and those of
12:
11:
5:
630:
628:
620:
619:
609:
608:
605:
604:
582:
581:
522:
511:www.vlfeat.org
498:
489:
480:
471:
444:
428:
397:
395:
392:
391:
390:
383:
380:
359:
356:
338:
335:
333:
330:
310:
296:
295:
280:
276:
271:
265:
262:
256:
248:
241:
237:
232:
228:
225:
220:
216:
212:
207:
203:
199:
196:
184:is defined as
177:
174:
170:log-likelihood
160:to log P(
150:
149:
138:
135:
131:
127:
124:
121:
118:
115:
110:
106:
102:
97:
93:
70:
67:
65:
62:
61:
60:
57:
24:, named after
13:
10:
9:
6:
4:
3:
2:
629:
618:
615:
614:
612:
603:
601:
597:
596:0-521-78019-5
593:
589:
585:
584:
577:
573:
568:
563:
559:
555:
550:
545:
541:
537:
533:
526:
523:
512:
508:
502:
499:
493:
490:
484:
481:
475:
472:
464:
457:
456:
448:
445:
441:
438:
432:
429:
426:
422:
419:
415:
411:
407:
406:Tommi Jaakola
402:
399:
393:
389:
386:
385:
381:
379:
377:
373:
368:
365:
357:
355:
353:
349:
345:
336:
331:
329:
327:
323:
278:
274:
269:
263:
260:
246:
239:
235:
230:
226:
218:
214:
210:
205:
201:
194:
187:
186:
185:
183:
182:Fisher kernel
176:Fisher kernel
175:
173:
171:
167:
163:
159:
155:
133:
125:
119:
116:
113:
108:
100:
95:
91:
83:
82:
81:
80:, defined as
79:
78:
68:
63:
58:
55:
54:
53:
51:
47:
43:
39:
34:
31:
27:
26:Ronald Fisher
23:
22:Fisher kernel
19:
598:
587:
539:
535:
525:
514:. Retrieved
510:
501:
492:
483:
474:
463:the original
454:
447:
439:
431:
409:
401:
369:
361:
340:
332:Applications
299:
297:
181:
179:
165:
161:
157:
153:
151:
74:
72:
69:Fisher score
35:
21:
15:
348:Naive Bayes
516:2017-03-04
324:being the
64:Derivation
40:(like the
602:SVM Book)
558:1932-6203
261:−
168:) is the
134:θ
117:
109:θ
105:∇
611:Category
576:28234999
536:PLOS ONE
425:Citeseer
382:See also
372:features
328:matrix.
567:5325198
75:Fisher
594:
574:
564:
556:
416:
344:tf–idf
48:(like
20:, the
466:(PDF)
459:(PDF)
298:with
152:with
77:score
592:ISBN
572:PMID
554:ISSN
414:ISBN
350:and
180:The
562:PMC
544:doi
114:log
52:):
16:In
613::
570:.
560:.
552:.
540:12
538:.
534:.
509:.
423:,
421:PS
354:.
346:,
599:(
578:.
546::
519:.
309:I
279:j
275:X
270:U
264:1
255:I
247:T
240:i
236:X
231:U
227:=
224:)
219:j
215:X
211:,
206:i
202:X
198:(
195:K
166:θ
164:|
162:X
158:θ
154:θ
137:)
130:|
126:X
123:(
120:P
101:=
96:X
92:U
Text is available under the Creative Commons Attribution-ShareAlike License. Additional terms may apply.