ModelBased Object Recognition Techniques Survey

1
Model-Based Object Recognition Techniques Survey
2
Preface

• Object recognition
• Classification
• Searching
• Model-based approach
• Searching process Single Model, dense image
• Algorithms and Techniques
• Pattern recognition
• Computer vision

3
Outline

• Pattern recognition
• Computer vision
• Process Analysis
• Data representation
• Decision making
• Conclusion
• Key references (review papers)

4
Outline

• Pattern recognition
• Definition and objective
• Algorithm development
• Approaches
• Computer vision
• Process Analysis
• Data representation
• Decision making
• Conclusion
• Key references (review papers)

5
Definition and objective1

• Pattern
• As opposite of a chaos.
• Pattern recognition
• The study of how machine can observe the
  environment, learn to distinguish patterns of
  interest from their background.
• Applications
• data mining, document classification, financial
  forecasting, organization and retrieval of
  multimedia databases, and biometrics.

6
Algorithm development1

• Three aspects of the design of a pattern
  recognition system
• data acquisition and preprocessing
• the choice of sensor, preprocessing technique
• data representation
• representation scheme
• decision making.
• the decision making model.

7
Approaches1

• The four best known approaches for pattern
  recognition
• Template matching
• Statistical classification
• Syntactic or structure matching
• Neural network.

8
Pattern recognition approaches1
9
Outline

• Pattern recognition
• Computer vision
• Preface
• Model-based vision system
• Conditions and Property Demand
• Process Issue
• Process Analysis
• Data representation
• Decision making
• Conclusion
• Key references (review papers)

10
Preface2

• Computer vision system
• To interpret the given visual data and to use
  the interpretation to complete a task.
• Typical tasks
• the navigation of autonomous vehicles
• the assembly or inspection of manufactured parts
  
• the analysis of microscopic images and medical
  x-rays.

11
Model-based vision system2

• Model
• The system have full knowledge of the shape of
  the desired object.
• Model-based
• To identify and locate a specified object(model)
  in the scene.
• Model-based vision system
• To identity the exact location and orientation of
  object model

12
Conditions and Property Demand2

• Process conditions and assumptions
• arbitrary or complicated shape
• Viewed from any direction
• Partially occluded by other objects.
• Property Demand from conditions
• Translation, rotational, scale invariance
• Robustness and stability
• Complexity
• Computing efficiency

13
Process Issue2

• Issues
• Type of sensor for data collection
• Resolution, precision, illumination
• Methods of constructing the necessary object
  model
• Representations of models of priori knowledge
• Means of describing the collected data and the
  models
• Descriptions of the collected data and the object
  models
• Methods of matching the object descriptions
• Matching strategies.

14

• General paradigm in model-based computer vision2

On Line
Data Collection
Low-level Processing
Data description
Matching
CAD description
Model analyzer
Model
Off Line
15
Outline

• Pattern recognition
• Computer vision
• Process Analysis
• Standard process
• Process analysis and classification
• Data representation
• Decision making
• Conclusion
• Key references (review papers)

16
Standard process

• Model-based object recognition on-line process
• data acquisition and preprocessing
• The choice of sensor - resolution, precision
• Preprocessing technique noise, defect, prepare
  for feature
• data representation
• Feature extraction
• Descriptor
• decision making
• Matching process

17
Process analysis and classification

• Assumption given image data
• Ignore data acquisition problem
• Preprocessing and Data representation
• Interaction issues
• Data representation and Decision making
• Low dependence
• Main tasks

18
Outline

• Pattern recognition
• Computer vision
• Process Analysis
• Data representation
• Definition
• Development and classification
• Decision making
• Conclusion
• Key references (review papers)

19
Definition

• Data representation method
• Feature
• A simple geometric characteristic or physical
  quality of the object model
• Descriptor
• Feature sets to represent a specific local or
  global area
• Preprocessing involvement
• Suited by selection of data representation

20
Development and classification3

• Data representation ? Shape analysis
• Shape analysis development
• Boundary Scalar Transform Techniques
• Boundary Space Domain Techniques
• Global Scalar Transform Techniques
• Global Space Domain Techniques

21
Boundary Scalar Transform Techniques3

• To be described indirectly by means of a 1-D
  charact-eristic function of the boundary instead
  of the 2-D boundary.
• From 2-D Shape to 1-D boundary representation
• tangent angle v.s. arc length (turning function)
• Fourier transform of boundary
• Bending energy
• Stochastic methods
• Arc height method

22

• Advantage
• Translational, Rotational, Scale invariant,
• Robustness good
• Disadvantage
• computing efficiency and complexity bad,
• data quantity require
• Possible mis-matching
• Keywords
• Hough transform, Houdsdoff distance, distance
  transform, similarity transform,
  Line-feature-based, Turning function, Bending
  energy, Stochastic

23
Boundary Space Domain Techniques3

• Take shape boundary as input produce an image, a
  graph, or other non-scalar values
• Chain code
• Syntactic techniques
• Boundary approximations
• Scale-space techniques
• Boundary decomposition

24

• Advantage
• Translational, Rotational, Scale invariant
• Disadvantage
• Computing efficiency and complexity bad,
• Robustness bad,
• Preprocessing require
• Keywords
• Vertices, edges, chain code, shape number,
  contour coding, syntactic, structural,
  scale-space, graph matching, oriented edge, pose
  clustering

25
Global Scalar Transform Techniques3

• The methods classified here compute a scalar
  result based on the global shape.
• Moments
• Shape matrices and vectors
• Morphological methods
• Keywords
• Moment inertial, morphological, distance metric,
  shape comparison, feature vector

26

• Advantage
• mathematically concise
• Disadvantage
• Translational, Rotational, Scale variant
• Optimization require
• Computing efficiency bad and high complexity
• Robustness and bad
• Preprocessing require .

27
Global Space Domain Techniques3

• Global space-domain methods are based on the
  analysis of the global shape (no scalar).
• Medial axis transform
• Shape decomposition
• Keywords
• Topological, nearest feature transform, spread
  bit map, unified transform, segmentation, Shape
  decomposition, convexity, curve segmentation

28

• Advantage
• Translational, Rotational, Scale invariant
• connectivity preservation
• single pixel width
• Disadvantage
• Optimum require
• robustness bad
• preprocessing require

29
Outline

• Pattern recognition
• Computer vision
• Process Analysis
• Data representation
• Decision making
• Definition
• Development and classification
• Conclusion
• Key references (review papers)

30
Definition

• Objective
• Develop the methods to identify the location and
  orientation of object model precisely and
  efficiently.
• Data representation involvement
• Adaptive application of selected feature
• Limitation of feature character

31
Development and classification

• Decision making ?Pattern matching
• Model-based pattern matching Development
• Template matching
• Structure pattern matching

32
Template matching1

• To determine the similarity between two entities
  of the same type. A template or prototype of the
  pattern to be recognized is available.
• Cross correlation
• Distance measurement
• Heuristic approach
• Window matching
• Keywords
• Template matching, object detection, vertices,
  edges, similarity measure, distance measure

33

• Advantage
• Translational invariant, intuitive implementation
• Disadvantage
• Rotational and scale variant
• Computing efficiency bad and high complexity
• Robustness bad, failure in occluded case
• Preprocessing require

34
Structure pattern matching1,4

• To identity object model by rule or grammar
  checking.
• Topological matching
• Indexing searching
• String matching
• Curve matching
• Keywords
• syntactic, structural, graph matching, feature
  indexing, hierarchical searching, heuristic
  search, tree searching, model database, occlusion
  analysis, string matching, curve matching

35

• Advantage
• Translational, Rotational, low complexity, good
  Robustness occluded scene invariant
• Disadvantage
• Scale variant
• Computing efficiency not good
• Preprocessing require
• False in dense image

36
Outline

• Pattern recognition
• Computer vision
• Process Analysis
• Data representation
• Decision making
• Conclusion
• Key references (review papers)

37

• Between preprocessing and data representation
• Existing inessential for preprocessing
• Exist for the demands of data representation
• Between data representation and decision making.
• Limitation and linking
• Interaction and Correlation (Relativity)
• Source of creative
• Essential data description
• Decision making process
• Methods assembly
• Process space

38
Data Representation
Boundary Scalar Transform
Boundary Space Domain
Global Scalar Transform
Global Space Domain
Feature based
Descriptor
Methods assembly
Process space
Decision making
Template matching
Structure matching
39
Outline

• Pattern recognition
• Computer vision
• Process Analysis
• Data representation
• Decision making
• Conclusion
• Key references (review papers)

40
Review papers

• 1A. K. Jain, P. W. Duin and J. Mao,
  Statistical pattern recognition A review,
  IEEE Transactions on Pattern Analysis and Machine
  Intelligence, Vol. 22, No. 1, pp. 4-37, Jan. 2000
• 2J. K. Aggarwal, A Model-based Object
  Recognition In Dense Range Image, ACM Computing
  Survey, Vol 25,No.1, pp. 5-43, Mar. 1995
• 3S. Loncaeic, A survey of shape analysis
  techniques, Pattern Recognition, Vol. 31, No. 8,
  pp. 9831001, 1998
• 4F. Stein and G. Medioni, Structural
  indexing-efficient 2D object recognition, IEEE
  Transactions on Pattern Analysis and Machine
  Intelligence, Vol. 14, No. 12, pp. 1198-1204,
  Dec. 1992