1 / 79

DIGITAL IMAGE PROCESSING

DIGITAL IMAGE PROCESSING. Dr J. Shanbehzadeh S . S. Nobakht. DIGITAL IMAGE PROCESSING. Chapter 12 – Object Recognition. Dr J. Shanbehzadeh Shanbehzadeh@gmail.com S . S. Nobakht. Road map of chapter 8. 12.3. 12.2. 12.2. 12.1. 12.3. 12.1. Some Basic Compression Methods.

eli
Download Presentation

DIGITAL IMAGE PROCESSING

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. DIGITAL IMAGE PROCESSING • Dr J. Shanbehzadeh • S. S. Nobakht ( J.Shanbehzadeh S. S. Nobakht)

  2. DIGITAL IMAGE PROCESSING Chapter 12 – Object Recognition • Dr J. Shanbehzadeh • Shanbehzadeh@gmail.com • S. S. Nobakht ( J.Shanbehzadeh S. S. Nobakht)

  3. Road map of chapter 8 12.3 12.2 12.2 12.1 12.3 12.1 • Some Basic Compression Methods • Structural Methods • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods • Patterns and Pattern Classes ( J.Shanbehzadeh S. S. Nobakht)

  4. 12.1 Patterns and Pattern Classes ( J.Shanbehzadeh S. S. Nobakht)

  5. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Recognition: Background Objects ( J.Shanbehzadeh S. S. Nobakht)

  6. Patterns and Pattern Classes Recognition: • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Human have developed a highly sophisticated skills for sensing their surroundings and recognizing what they observed, e.g., distinguishing insect from leaves, understanding spoken words, recognizing a face. We would like to develop a machine having similar capability. ( J.Shanbehzadeh S. S. Nobakht)

  7. Patterns and Pattern Classes Pattern Recognition? • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods • A pattern is a set of consistent, characteristic form, style of an object, each individual or group, such as • Fingerprints • Pattern of queen’s behavior • Color • shape • Pattern recognition is a study of how machines can • Observe its environments • Learn to distinguish patterns of interest • Make sound decisions and reasonable assignments of patterns to possible classes. Pattern class is a family of patterns that share some common properties. ( J.Shanbehzadeh S. S. Nobakht)

  8. Patterns and Pattern Classes Pattern Recognition • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods • A abstract representation of pattern recognition system. World Observations Class Space Pattern Space ( J.Shanbehzadeh S. S. Nobakht)

  9. Patterns and Pattern Classes Pattern Recognition • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods • “Sorting incoming Fishes on a conveyor according to species using optical sensing”. Sea bass Species Salmon ( J.Shanbehzadeh S. S. Nobakht)

  10. Patterns and Pattern Classes Problem Analysis • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods • What are steps in the process? • What would cause problems during the sensing? • What kind of information can distinguish one species from the other? • Taking pictures • Isolating a fish • Taking some measurements • Classifying it. ( J.Shanbehzadeh S. S. Nobakht)

  11. Patterns and Pattern Classes Problem Analysis • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods • What would cause problems during the sensing? • Lighting, position of the fish, etc. • What kind of information can distinguish one species from the other? • Suggested features: length, lightness, width, number and shape of fins, etc… • Prior knowledge: • Sea bass is generally longer than salmon. This gives us a tentative model. ( J.Shanbehzadeh S. S. Nobakht)

  12. Patterns and Pattern Classes Pattern Recognition System • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods ( J.Shanbehzadeh S. S. Nobakht)

  13. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Pattern Recognition System • Preprocessing • Use a segmentation operation to isolate fishes from one another and from the background • May apply some noise reductions prior the segmentation. • Information from a single fish is sent to a feature extractor whose purpose is to reduce the data by measuring certain features. ( J.Shanbehzadeh S. S. Nobakht)

  14. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Pattern Recognition System • The features are passed to a classifier • Select the length of the fish as a possible feature for discrimination • Classification • Design which class the fish is based on the selected features and its threshold value. How Can We Choose The Threshold ? ( J.Shanbehzadeh S. S. Nobakht)

  15. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Selecting Features • Features are any extractable measurements used. • Often contains noises, which leads to misclassification errors. • Feature selection is then required to choose and extract features that • - Are computationally feasible • - Lead to good pattern recognition success. • -Reduce the problem data (e.g. raw measurements) into manageable amount of information without discarding valuable information. ( J.Shanbehzadeh S. S. Nobakht)

  16. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Selecting Features • From training data, we obtained histogram of the length of the two types of fishes. • How to choose a threshold value that can make reliable decision? • -The length is a poor feature alone! ( J.Shanbehzadeh S. S. Nobakht)

  17. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Selecting Features • Try out another feature, average lightness of the fishes. • It is much better separate, so it is easier to choose the threshold value. • However, can we obtain a perfect decision? ( J.Shanbehzadeh S. S. Nobakht)

  18. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Cost of Error • Moving decision boundary affects cost of error. • Another prior knowledge • Salmon is a lot more expensive than sea bass. • Customer who buy salmon will object vigorously if they see sea bass in their cans. • Customers who buy sea bass will be happy if they occasionally see some expensive salmon in their can. • How does this knowledge affect our decision? ( J.Shanbehzadeh S. S. Nobakht)

  19. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Threshold decision boundary and cost relationship • Move our decision boundary toward smaller values of lightness in order to minimize the cost (reduce the number of sea bass that are classified salmon!) • Task of decision theory ( J.Shanbehzadeh S. S. Nobakht)

  20. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Multiple Features • Adopt the lightness and add the width of the fish • Each fish is now represented as a point using feature vector, xT. Fish Width Lightness ( J.Shanbehzadeh S. S. Nobakht)

  21. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Multiple Features • Features are often arranged into a d-dimensional feature space. A plot of feature vectors into feature space is called scatter plot. ( J.Shanbehzadeh S. S. Nobakht)

  22. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Decision Boundary • Can we do better using another decision theory? • How about using more complex model for more complex boundary? ( J.Shanbehzadeh S. S. Nobakht)

  23. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Performance to Unknown Sample • However, our satisfaction is premature because the central aim of designing a classifier is to correctly classify novel input? • Issue of generalization ! ( J.Shanbehzadeh S. S. Nobakht)

  24. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Performance to Unknown Sample ( J.Shanbehzadeh S. S. Nobakht)

  25. Patterns and Pattern Classes Training Data Physical Environment • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Data Acquisition / Sensing Pre-Processing Pre-Processing Feature Extraction / Selection Feature Extraction Features Features Model Learning / Estimation Model Classification Post-Processing Decision ( J.Shanbehzadeh S. S. Nobakht)

  26. Patterns and Pattern Classes Start • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Collect Data The Design Cycle Choose Features • Data Collection • Feature Choice • Model Choice • Training • Evaluation • Computational Complexity Prior Knowledge Choose Model Train Classifier Evaluate Classifier End ( J.Shanbehzadeh S. S. Nobakht)

  27. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods The Design Cycle • Data Collection • Is the collected data large enough to represent the system? • - To assume good performance of the system, large data is required. • Feature Choice • Is the collected data large enough to represent the system • -Simple to extract, invariant to irrelevant transformation insensitive to noise. ( J.Shanbehzadeh S. S. Nobakht)

  28. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods The Design Cycle • Model Choice • Parametric or non-parametric model • Template, decision-theoretic, neural, structural or hybrid model. • Training • Training is a process of using data to determine the classifier. Many different procedures for training classifiers. ( J.Shanbehzadeh S. S. Nobakht)

  29. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Computational Complexity • Evaluation • Measure the error rate (or performance) of the system. • This to measure the system performance and to identify the need for improvements in its components. • Computation Complexity • What is the trade-off between computational ease and performance? • (How an algorithm scales as a function of the number of features, patterns or categories? ) ( J.Shanbehzadeh S. S. Nobakht)

  30. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Three Pattern Recognition Approaches • Statistical PR • Syntactic PR • Neural PR ( J.Shanbehzadeh S. S. Nobakht)

  31. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Statistical PR • May be called decision-theoretic PR • Classifying is based on statistic basis. • Underlying model is of a state of probabilities and/or probability density function. ( J.Shanbehzadeh S. S. Nobakht)

  32. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Statistical PR • Classifying is based on structural information. • Underlying model is of defining primitive and a set of grammars or rules. Class 1 Structure Class 2 Structure Class N Structure . . . Structural Analysis Structural Matcher Relevant Match ( J.Shanbehzadeh S. S. Nobakht)

  33. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Statistical PR ( J.Shanbehzadeh S. S. Nobakht)

  34. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Statistical PR ( J.Shanbehzadeh S. S. Nobakht)

  35. Patterns and Pattern Classes • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Neural PR • Classification is done based on knowledge of how biological neural system store and manipulate information. ( J.Shanbehzadeh S. S. Nobakht)

  36. Patterns and Pattern Classes Comparison • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods ( J.Shanbehzadeh S. S. Nobakht)

  37. 12.2 Recognition Based on Decision-Theoretic Methods ( J.Shanbehzadeh S. S. Nobakht)

  38. Matching Matching • Optimum Statistical Classifiers • Neural Networks • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods ( J.Shanbehzadeh S. S. Nobakht)

  39. Matching • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods • Pattern : Mathematical Representation • X is Pattern Features Vector • Pattern classes is a family of patterns that share some common properties • Patter classes is denoted by 𝜔_1, 𝜔_2, …, 𝜔_𝑀 where M is the number of classes ( J.Shanbehzadeh S. S. Nobakht)

  40. Matching • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Decision Theory • Discriminant Function • x is n-dimentional pattern vector • Decision rule for class 𝜔_𝑖 ( J.Shanbehzadeh S. S. Nobakht)

  41. Matching • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Decision Theory ( J.Shanbehzadeh S. S. Nobakht) Linearly Separable

  42. Matching • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Decision Theory ( J.Shanbehzadeh S. S. Nobakht) Nonlinearly Separable

  43. Matching • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Decision Theory ( J.Shanbehzadeh S. S. Nobakht) Non-Separable

  44. Matching • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Minimum Distance Classifier ( J.Shanbehzadeh S. S. Nobakht)

  45. Matching • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Minimum Distance Classifier ( J.Shanbehzadeh S. S. Nobakht)

  46. Matching • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Decision Theory ( J.Shanbehzadeh S. S. Nobakht)

  47. Matching • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Decision Theory For decision boundary classes and : ( J.Shanbehzadeh S. S. Nobakht)

  48. Matching • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Decision Theory ( J.Shanbehzadeh S. S. Nobakht) For decision boundary classes and : Decision Boundary

  49. Matching • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Decision Theory • If 𝑑_12 (𝑥)>0 , x belongs to class 𝜔_1 • If 𝑑_12 (𝑥)<0 , x belongs to class 𝜔_2 ( J.Shanbehzadeh S. S. Nobakht)

  50. Matching • 12.1 Patterns and Pattern Classes • 12.2 Recognition Based on Decision-Theoretic Methods • 12.3 Structural Methods Decision Theory ( J.Shanbehzadeh S. S. Nobakht)

More Related