Contextual Image Search

30
Contextual Image Search Wenhao Lu Wenhao Lu , Jingdong Wang , Xian-Sheng Hua, Shengjin Wang , Shipeng Li , Jingdong Wang , Xian-Sheng Hua, Shengjin Wang , Shipeng Li Tsinghua University, Beijing, P. R. China, Tsinghua University, Beijing, P. R. China, Microsoft Research Asia, Beijing, P. R. China, Microsoft Research Asia, Beijing, P. R. China, MM 2011

description

Contextual Image Search. Wenhao Lu , Jingdong Wang , Xian- Sheng Hua , Shengjin Wang , Shipeng Li Tsinghua University, Beijing, P. R. China, Microsoft Research Asia, Beijing, P. R. China,. MM 2011. Outline. System overview Database construction - PowerPoint PPT Presentation

Transcript of Contextual Image Search

Page 1: Contextual Image Search

Contextual Image SearchContextual Image Search

Wenhao LuWenhao Lu , Jingdong Wang , Xian-Sheng Hua, Shengjin Wang , Shipeng Li , Jingdong Wang , Xian-Sheng Hua, Shengjin Wang , Shipeng Li

Tsinghua University, Beijing, P. R. China, Tsinghua University, Beijing, P. R. China,

Microsoft Research Asia, Beijing, P. R. China, Microsoft Research Asia, Beijing, P. R. China,

Wenhao LuWenhao Lu , Jingdong Wang , Xian-Sheng Hua, Shengjin Wang , Shipeng Li , Jingdong Wang , Xian-Sheng Hua, Shengjin Wang , Shipeng Li

Tsinghua University, Beijing, P. R. China, Tsinghua University, Beijing, P. R. China,

Microsoft Research Asia, Beijing, P. R. China, Microsoft Research Asia, Beijing, P. R. China,

MM 2011

Page 2: Contextual Image Search

MM 2011

Outline

System overview Database construction Contextual image search with text/image input Experiment Future Work

2

Page 3: Contextual Image Search

MM 2011

System overview

3

Text input

Page 4: Contextual Image Search

MM 20114

Image input

System overview

Page 5: Contextual Image Search

MM 20115

Database construction

Page 6: Contextual Image Search

MM 20116

Database construction

1. Feature extraction (MSER)

extracts stable regions from the image by considering the change in area w.r.t the change in intensity of a connected component defined

Page 7: Contextual Image Search

MM 20117

Database construction

2. SIFT descriptor

Page 8: Contextual Image Search

MM 20118

Database construction

2. SIFT descriptor

Page 9: Contextual Image Search

MM 20119

Contextual Image Search WithText Input

1. Context Capturing

visual contexts: vision-based page segmentation algorithm (VIPS)

textual contexts: page title / document title local context

Page 10: Contextual Image Search

MM 201110

vision-based page segmentation

Traditional DOM tree

Page 11: Contextual Image Search

MM 201111

vision-based page segmentation

VIPS

Page 12: Contextual Image Search

MM 201112

vision-based page segmentation

Tag cue: <HR>Color cue: background colorText cueSize cue

DOM tree +Visual Info

Page 13: Contextual Image Search

13

Contextual Image Search WithText Input

2. Contextual Query Augmentation

Goal: remove possible ambiguities Augmented query = query + textual context

Candidate augmented query

evaluate the relevance betweenthe context and augmented query (Okapi BM25)MM 2011

Page 14: Contextual Image Search

14

MM 2011

2. Contextual Query Augmentation

: extended context (using synonyms, stemming, and so on)

k=2.0, b=0.75

Okapi BM25

~

Contextual Image Search WithText Input

Page 15: Contextual Image Search

2. Contextual Query Augmentation

Rank score =

: static score (ex. the Web page holding this image)

3. Image Search by Text

15

Contextual Image Search WithText Input

Page 16: Contextual Image Search

MM 2011

Contextual Reranking

textually contextual reranking

visually contextual reranking

, : discarding the augmented query related words

1. Filter out images whose semantic contents may not be relevant to the query. (compute local textual context and query)

16

Page 17: Contextual Image Search

MM 2011

Contextual Reranking visually contextual reranking

2. Visual word weight:

Find common pattern

3. Compute similarity

:visual contexts

: an image

: histogram vector of i

: histogram vector of k 17

Page 18: Contextual Image Search

MM 2011

Overall Ranking

= 0.2

= 0.2

=1

18

Page 19: Contextual Image Search

MM 2011

Contextual Image Search with Image Input

3

1. Search to annotation

discovers the candidate textual queries using the technique “Annotating images by mining search result” (IEEE 2008)

19

Page 20: Contextual Image Search

MM 2011

Contextual Image Search with Image Input

3

1. Search to annotation

20

Page 21: Contextual Image Search

MM 2011

Contextual Image Search with Image Input

3

1. Search to annotation

First : find similar image

Second: surrounding texts of the obtained duplicated images are mined to get a list of candidate textual queries

visual features

semantic features

Page 22: Contextual Image Search

MM 2011

Contextual Image Search with Image Input

1. Search to annotation

22

Page 23: Contextual Image Search

MM 2011

Contextual Image Search with Image Input

2. Contextual query identification

calculate ~

23

Page 24: Contextual Image Search

MM 2011

Experiment

24

15,000,000 images and associated web pages

5 users (level 0~level 3)

Page 25: Contextual Image Search

MM 2011

Experiment

25

0.95

0.65

nDCG curves

Page 26: Contextual Image Search

MM 2011

Experiment

26

Visual Result for Text Input

Page 27: Contextual Image Search

MM 2011

Experiment

27

Visual Result for Text Input (Textual Reranking)

Page 28: Contextual Image Search

MM 2011

Experiment

28

Visual Result for Text Input (Visual Reranking)

Page 29: Contextual Image Search

MM 2011

Experiment

29

Visual Result for Image Input

textual query “Van gogh”

Page 30: Contextual Image Search

MM 2011

Future Work

30

1. More general contextual image search, including mobile image search with wider contexts (e.g., position, time, and history)

2. Extend contextual image search to contextual video search by applying the proposed methodology and investigating extra video contexts