CS 556: Computer Vision. Lecture 18

Size: px

Start display at page:

Download "CS 556: Computer Vision. Lecture 18"

Toby Albert Fowler
5 years ago
Views:

1 CS 556: Computer Vision Lecture 18 Prof. Sinisa Todorovic 1

2 Color 2

3 Perception of Color The sensation of color is caused by the brain Strongly affected by: Other nearby colors Adaptation to previous views State of mind 3

4 Do the Gray Curves Have the Same Color? 4

5 Do the Gray Curves Have the Same Color? 5

6 Image Features -- Color Trichromacy: Color = r Red + g Green + b Blue Biologically justified: 3 types of color receptors = Cones In general, Color = a A + b B + c C The choice of A, B, C Color description system 6

7 RGB Color Space 7

8 Color Channels original red green blue image channel channel channel 8

9 RGB Color Space Inconvenient to represent intuitive notions of hue, saturation, and brightness (value) Does not capture human intuitions that hues form a circle 9

10 HSV Color Space 10

11 HSV Color Space Hue and Saturation 11

12 HSV Color Space 12

13 RGB to HSI Conversion H =, if B G 360, if B>G 1 = cos 1 [(R G)+(R B)] 2 [(R G) 2 +(R B)(G B)] 1/2 S =1 3 R + G + B min(r, G, B) I = 1 (R + G + B) 3 13

14 HSI to RGB Conversion Sector: 0 H 120 B = I(1 S) R = I 1+ S cos H cos(60 H) G =3I (R + B) 14

15 HSI to RGB Conversion Sector: 120 H 240 H = H 120 R = I(1 S) G = I 1+ S cos H cos(60 H) B =3I (R + G) 15

16 HSI to RGB Conversion Sector: 240 H 360 H = H 240 G = I(1 S) B = I 1+ S cos H cos(60 H) R =3I (B + G) 16

17 Issues with Using Color as Image Feature 17

18 Issues with Using Color as Image Feature Color at one pixel location is not very informative 17

19 Issues with Using Color as Image Feature Color at one pixel location is not very informative Color may characterize an image region 17

20 Issues with Using Color as Image Feature Color at one pixel location is not very informative Color may characterize an image region Color constancy -- subjective color perception 17

21 Issues with Using Color as Image Feature Color at one pixel location is not very informative Color may characterize an image region Color constancy -- subjective color perception 17

22 Perceptual Grouping 18

23 What do you see? 19

24 Gestalt Laws Arise from constraints of the real world Similarity Proximity Closure Good continuation Common fate Figure-ground Symmetry Periodicity 20

25 Gestalt Laws -- Similarity We group image parts with similar properties (color, shape, texture) Because same matter in the real world often yields same image properties 21

26 Gestalt Laws -- Proximity We tend to group nearby image parts Because matter is cohesive resulting in meaningful configurations of nearby objects 22

27 Gestalt Laws -- Closure We tend to ignore gaps and hallucinate complete, closed contours Because objects have closed surfaces, and so their image projections should have closed boundaries 23

28 Gestalt Laws -- Good Continuation We prefer to see configurations forming smooth contours Because objects have locally smooth surfaces 24

29 Gestalt Laws -- Common Fate We tend to see distinct image parts with same motion as a unit Because parts of a moving object move coherently in the same manner 25

30 Parallelism, Symmetry α 1 r q 1 r r q 2 n 1 n 2 α 2 two parallel contours two symmetric contours 26

31 Gestalt Laws -- Figure-Ground We see certain image areas as foreground or figure and the remaining areas as background or ground Because of a target oriented nature of human vision 27

32 Gestalt Laws -- Symmetry We tend to see symmetric image parts as figure Because many objects are symmetric due to functionality/growth/reproduction processes 28

33 Gestalt Laws -- Periodicity We tend to see objects in image parts that spatially repeat Because many objects are spatially periodic due to functionality/growth/reproduction processes 29

34 Issues How to formalize Gestalt laws? What Gestalt law should we apply first? Different orderings Different groupings 30

35 Top-down Interpretation from Context Context and feedback from higher levels may resolve low-level ambiguities Because of the context we may not be even aware of alternatives 31

36 Top-down Interpretation from Context Context and feedback from higher levels may resolve low-level ambiguities Because of the context we may not be even aware of alternatives 31

37 Top-down Interpretation from Context Context and feedback from higher levels may resolve low-level ambiguities Because of the context we may not be even aware of alternatives 31

38 Top-down Interpretation from Context Context and feedback from higher levels may resolve low-level ambiguities Because of the context we may not be even aware of alternatives 31

39 Top-down Interpretation from Context Context and feedback from higher levels may resolve low-level ambiguities Because of the context we may not be even aware of alternatives 31

CS 534: Computer Vision Segmentation and Perceptual Grouping

CS 534: Computer Vision Segmentation and Perceptual Grouping Spring 2005 Ahmed Elgammal Dept of Computer Science CS 534 Segmentation - 1 Where are we? Image Formation Human vision Cameras Geometric Camera