1. Click the Occluded Shape
Supervisor:
Prof. Ohad Ben Shahar
Click the Occluded Shape
Ran Arieli
Bar Harel

2. 1
Introduction

3. Introduction
A major challenge for visual recognition is to describe shapes flexibly enough to allow generalization over different views.

4. Introduction
For decades, an influential solution championed by computer vision research has been to describe shapes according to a type of underlying skeletal representation, a leading candidate for which is a geometric structure known as the medial axis (Blum, 1973).

5. 2
Medial Axis

6. The medial axis of an object is the set of all points having more than one closest point on the object's boundary. 

7. Medial Axis Examples
Here is an example of
A shape –
A simple rectangle.
Which points have more than one closest point on the object’s boundary?

8. Medial Axis Examples
Here is the same rectangle, this time with its medial axis colored in red.

9. Medial Axis Examples
Another shape, this time a triangle.
Which points construct its medial axis?

10. Medial Axis Examples
The triangle with its medial axis.

11. Medial Axis Examples
Not only simple geometric shapes have a medial axis, for example, let’s find this bunny’s medial axis.

12. Medial Axis Examples
The bunny with its medial axis.
And a carrot.

13. “Please Tap the Shape, Anywhere You Like”3
Yale research article
“Please Tap the Shape, Anywhere You Like”
2014, by Chaz Firestone and Brian J. Scholl

14. Yale research article 1. Concept
Attempting to figure out the role and significance of medial axis skeletonization in human vision.
2. Experiment
The experiment took place at Times Square, as pedestrians were presented a tablet showing a shape and were asked to touch the shape “anywhere they wished”.
3. Results
The experiment produced unusually direct evidence that human vision represents shapes in a skeletal format –
When participants simply touched a shape anywhere they pleased, their chosen locations formed the shape’s medial-axis skeleton

15. Yale research article 1. Concept
Attempting to figure out the role and significance of medial axis skeletonization in human vision.
2. Experiment
The experiment took place at Times Square, as pedestrians were presented a tablet showing a shape and were asked to touch the shape “anywhere they wished”.
3. Results
The experiment produced unusually direct evidence that human vision represents shapes in a skeletal format –
When participants simply touched a shape anywhere they pleased, their chosen locations formed the shape’s medial-axis skeleton

16. Yale research article 2. Experiment
The experiment took place at Times Square, as pedestrians were presented a tablet showing a shape and were asked to touch the shape “anywhere they wished”.

17. Yale research article 1. Concept
Attempting to figure out the role and significance of medial axis skeletonization in human vision.
2. Experiment
The experiment took place at Times Square, as pedestrians were presented a tablet showing a shape and were asked to touch the shape “anywhere they wished”.
3. Results
The experiment produced unusually direct evidence that human vision represents shapes in a skeletal format –
When participants simply touched a shape anywhere they pleased, their chosen locations formed the shape’s medial-axis skeleton

18. Yale research article 1. Concept
Atemping to figure out the role and significance of medial axis skeletonization in human vision.
2. Experiment
The experiment took place at Times Square, as pedestrians were presented a tablet showing a shape and were asked to touch the shape “anywhere they wished”.
3. Results
The experiment produced unusually direct evidence that human vision represents shapes in a skeletal format –
When participants simply touched a shape anywhere they pleased, their chosen locations formed the shape’s medial-axis skeleton

19. 4
Our experiment

20. Our experiment
Our experiment was designed to answer two goals:
1. Strengthen Yale experiment.
2. Examine the significance
of medial axis on human vision r regarding hidden shapes.

21. Our experiment
Our experiment was designed to answer two goals:
1. Strengthen Yale experiment.
2. Examine the significance
of medial axis on human vision r regarding hidden shapes.

22. Our experiment
Our experiment was designed to answer two goals:
1. Strengthen Yale experiment.
2. Examine the significance
of medial axis on human vision r regarding hidden shapes.

23. ? Our experiment Our experiment was designed to answer two goals:
1. Strengthen Yale experiment.
2. Examine the significance
of medial axis on human vision r regarding hidden shapes.

24. Experiment description
Our experiment took place at many locations in Be’er Sheva.
We approached random people, presented them with the tablet and asked them to touch a series of shapes, anywhere they would like.

25. Experiment description
Some of the shapes were occluded.

26. Experiment description
Some of the shapes were occluded.
Some were not.

27. Experiment description
Some of the shapes were occluded.
Some were not.
There were in total 1000 participants, which tapped each of the 15 shapes.

28. Analyzing non-occluded shapes

29. Analyzing non-occluded shapes

30. Analyzing non-occluded shapes

31. Analyzing non-occluded shapes

32. Analyzing non-occluded shapes
For each shape we calculated the average distance of the tapped points from the medial axis.
We generated 1000 sessions of 1000 randomly chosen points.
From these 1000 sessions we chose the one with minimum average distance.

33. Analyzing non-occluded shapes
For each shape we calculated the average distance of the tapped points from the medial axis.
We generated 1000 sessions of 1000 randomly chosen points.
From these 1000 sessions we chose the one with minimum average distance.

34. Analyzing non-occluded shapes

35. Is the average distance from taps to medial axis
11.12 Pixels
Is the average distance from taps to medial axis
Pixels
Is the average distance for the randomly chosen taps

36. Is the average distance from taps to medial axis
11.12 Pixels
Is the average distance from taps to medial axis
Pixels
Is the average distance for the randomly chosen taps
Ratio of 3.705!

37. Analyzing occluded shapes

38. Analyzing occluded shapes

39. Analyzing occluded shapes

40. Analyzing occluded shapes

41. Many taps don’t fit the medial axis!
Did medial axis lose its role in human vision when dealing with occluded shapes?

42. Many taps don’t fit the medial axis!
Did medial axis lose its role in human vision when dealing with occluded shapes?
Let’s try another approach.

43. Analyzing occluded shapes

44. Analyzing occluded shapes

45. Analyzing occluded shapes

46. Analyzing occluded shapes - old approach

47. Analyzing occluded shapes - old approach
37.14 Pixels
distance taps <-> medial axis

48. Analyzing occluded shapes - old approach
37.14 Pixels
distance taps <-> medial axis
56.18 Pixels
distance randomly chosen taps <-> medial axis

49. Analyzing occluded shapes - old approach
37.14 Pixels
distance taps <-> medial axis
56.18 Pixels
distance randomly chosen taps <-> medial axis
Ratio of 1.512

50. Analyzing occluded shapes - new approach

51. Analyzing occluded shapes - new approach
12.95 Pixels
distance taps <-> medial axis

52. Analyzing occluded shapes - new approach
12.95 Pixels
distance taps <-> medial axis
33.85 Pixels
distance randomly chosen taps <-> medial axis

53. Analyzing occluded shapes - new approach
12.95 Pixels
distance taps <-> medial axis
33.85 Pixels
distance randomly chosen taps <-> medial axis
Ratio of
2.612

54. Thanks!
Any questions?