temu kembali citra berbasis anotasi automatis · 12/21/2009 1 imam abu daud, yeni herdiyeni, sri...

Post on 01-May-2019

217 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

12/21/2009

1

IMAM ABU DAUD, YENI HERDIYENI, SRI NURDIATI,

DEPARTMENT OF COMPUTER SCIENCE BOGOR AGRICULTURAL UNIVERSITY

INDONESIA

INTRODUCTION BACKGROUND THE PROPOSED METHOD THE EXPERIMENTS & RESULTS EVALUATION CONCLUSIONS

MANUAL ANNOTATION ARE SUBJECTIVE AND TIME CONSUMING

4

19 October 2009

http://www.similar-images.googlelabs.com/

BUT IT TAKES LONGER TO FIND IMAGE THAN TEXT BASED

12/21/2009

2

Automatically Annotation

? ? ?

road, grass, sign

grass, mountain, cloud

grass, plant plant grass, road, house

grass, tree, water

grass, tree,water, mountain

grass, house

grass, road,

12/21/2009

3

Grid Segmentation

Statistical Machine

Translation (EM ALGORITHM)

2 X 3

4 X 6

8 X 12

road, grass, signTRANSLATION TABLE

( )NS

a a

r wE

n N n

Normalized Score

Bad Good

-1 10

12/21/2009

4

:

Statistical Machine

Translation

LSI

DATA CONSIST OF 1000 IMAGES

TRAINING (750 IMAGES)

▪ A : GRID 2 X 3

▪ B : GRID 4 X 6

▪ C : GRID 8 X 12

TESTING (250 IMAGES)

▪ A : GRID 2 X 3

▪ B : GRID 4 X 6

▪ C : GRID 8 X 12

12/21/2009

5

Anotasi Manual Anotasi Automatis

word1, word2,

69 word

LatentSemanticIndexing

Hasil temu kembali

Statistical Machine

Translation + EM Algorithm

Recall & Precision

clausa query : 0.544, text query : 0.251

The experimental results showed that latent semantic indexing with clause query works better than text query with average precision is 0.5442 and 0.2509 for clause and text query respectively.

The proposed method of latent semantic indexing succeeds to exploit semantic value of automatic-annotation-based image retrieval.

12/21/2009

6

top related