Color-Texture Analysis for Content-Based Image Retrieval

1
Color-Texture Analysis for Content-Based Image
Retrieval

• Anh-Minh Hoang (W03213684)
• Supervisor Vassilis Kodogiannis
• M.Sc. in Intelligent and Multi-Agent Systems,
• Harrow School of Computer Science.

2
Outline

• Introduction to the problem
• The goals of the work
• Introduction to the approach
• The relevance of the work to the areas of
  Intelligent Systems
• Related works
• Evaluation methods

3
Introduction

• The volume of digital image archives is growing
  rapidly and has become very large
• Large amount of visual data is available on
  digital libraries or on the WWW.
• The needs for searching visual information such
  as images, videos are emerging

4
Introduction (cont.)

• Manual image annotations can be used to a certain
  extent to help image search, but the feasibility
  of such approach to large databases is a
  questionable issue
• Content-based image retrieval (CBIR) aims at
  efficient retrieval of relevant images from large
  image databases based on automatically derived
  imagery features such as color, texture, shape

5
Introduction (cont.)
6
Goals

• To automatically derive color and texture feature
  from image
• To automatically partition an image into disjoint
  region coherently different in color and texture
  (image segmentation)
• To build an image retrieval system using color
  and texture information

7
Approach

• Color-texture measurement (see Minh A. Hoang et
  al, Signal Processing, pp. 265275, February
  2005)
• Multiscale Region-Boundary Refinement for
  Color-Texture Segmentation
• Features and regions indexing and matching for
  image retrieval

8
Color-texture Feature
9
Segmentation Multiscale Approach
10
Color-texture Segmentation
Ground truth
11
Color-texture Segmentation (cont.)
12
Image Retrieval System
13
Applications in some areas ofIntelligent Systems

• Robot vision, object recognition, object tracking
  (e.g. robot soccer, intelligent vehicles driver
  assistance) visual feature extraction and image
  segmentation is fundamental
• Search engines for visual information, automatic
  annotation of visual database, automatic
  detection of salient features

14
Related Works

• IBM QBIC, MIT Photobook, Columbia VisualSEEK and
  WebSEEK, PicToSeek, BlobWord image retrieval
  systems
• J. Malik et al, Contour and texture analysis for
  image segmentation, International Journal of
  Computer Vision 43(1), pp. 727, 2001
• J. Freixenet et al, Color Texture Segmentation
  by Region-Boundary Cooperation, in The Eighth
  European Conference on Computer Vision, pp.
  250261, Springer Verlag, (Prague, Czech
  Republic), may 2004.

15
Related Works (cont.)

• M. Tabb et al, Multiscale image segmentation by
  integrated edge and region detection, IEEE
  Trans. on Image Processing 6(5), pp. 642655,
  1997
• P. Schroeter et al, Hierarchical image
  segmentation by multi-dimensional clustering and
  orientation-adaptive boundary refinement,
  Pattern Recognition 28(5), pp. 695709, 1995.
• M. Mirmehdi and M. Petrou, Segmentation of color
  textures, IEEE Trans. on PAMI 22(2), pp.
  142159, 2000.
• A. W. M. Smeulders et al, Content-based image
  retrieval at the end of the early years, IEEE
  Trans. on PAMI 22(12), pp. 13491380, 2000.

16
Evaluation methods

• Evaluation of color-texture feature extraction
  and image segmentation based on
• Compare with ground truth samples (or with human
  segmentations)
• Compare to results from other works
• Verify by human perception (heuristics)

17
Evaluation methods (cont.)

• Evaluation of image retrieval system based on
• Average precision vs. number of retrieved images
  for several query types
• Average number of steps to get to desired results
  based on relevant feedbacks
• Heuristics (verify by human perception)