supporting landmark image retrieval with skyline extraction techniques

17
SUPPORTING LANDMARK IMAGE RETRIEVAL WITH SKYLINE EXTRACTION TECHNIQUES Date : 2012 / 04 / 12 資資資10077034 資資資 @ LAB603

Upload: oistin

Post on 22-Feb-2016

29 views

Category:

Documents


0 download

DESCRIPTION

Supporting Landmark Image Retrieval with Skyline Extraction Techniques . Date : 2012 / 04 / 12 資訊碩一 10077034 蔡勇儀 @ LAB603 . Outline. Introduction Preliminaries Method Experimental result Conclusions. Introduction. Image retrieval have more challenge than text retrieval. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

SUPPORTING LANDMARK IMAGE RETRIEVAL WITH SKYLINE

EXTRACTION TECHNIQUES Date : 2012 / 04 / 12資訊碩一 10077034 蔡勇儀 @ LAB603

Page 2: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Outline Introduction Preliminaries Method Experimental result Conclusions

Page 3: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Introduction Image retrieval have more challenge

than text retrieval.

Use low-level feature to identify objects is non-trivial.

Skyline is a good feature to identify landmark contour.

Page 4: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Outline Introduction Preliminaries Method Experimental result Conclusions

Page 5: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Traditional image retrieval Traditional techniques for image retrieval

rely on some metadata (i.e., keywords or captions of image)

It isn’t a good method for practical applications since it is labor intensive and need more space.

Page 6: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Content-based image retrieval Content-based image retrieval(CBIR) is

using image’s feature for retrieval.

Query-by-example Query-by-sketch

Relevance feedback technique

Page 7: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Identifying landmarks with skylines Region-growing-based

Using same color region extend find the skylines Weak robustness on the sky with cloud or

complex background

Edge-based Edge detection Retaining the points which are luminous intensity

change sharply Using other method to find clear skyline.

Page 8: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Outline Introduction Preliminaries Method Experimental result Conclusions

Page 9: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Method(1/4) - Difficulties Small portion of edge points constitute

the proper skyline.

Skyline may not fully-connected.

Steep or tortuous.

Page 10: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Method(2/4) - RON RON(radius of neighborhood)

Defining the search region (ex. RON = 2 -> grid zone = 5*5) Solving the difficulties.

Page 11: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Method(3/4) – Direction priority Using clockwise

Good direction for traversal skyline.

Page 12: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Method(4/4) – Find the skyline Using Sobel Filter for edge detection Using DFS for search. (Need decide the

start point?)

Page 13: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Outline Introduction Preliminaries Method Experimental result Conclusions

Page 14: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Experimental Result DP VS This method ( O(n2) & O(n) )

Page 15: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Outline Introduction Preliminaries Method Experimental result Conclusions

Page 16: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Conclusions This is a feasible for landmark image retrieval.

If there are more line cross the image?

If sky have complex object(ex. Light or cloud…)?

If the beginning point is not in skyline?

More problem need to solving…

Page 17: Supporting Landmark Image Retrieval with Skyline Extraction Techniques

Thank for your listening! Q&A