Week 5A: Shape perception

Shape perception: Successive processing stages

Local features, 2D, 2.5D, 3D

Who was the inventer of gestalt pscyhology

Max Wertheimer

Gestalt psychology: The whole is more than (…)

the sum of its parts

Five main laws of gestalt psychology

1. Proximity

2. Similarity

3. Closure

4. Good continuation

5. Common fates

Gestalt laws: 1,2,3,4,5,6

Proximity, similarity, closure, good continuation, common region, connectedness

Gestalt psychology: Good continuation

Contours that vary smoothly are grouped together (example of 2 horses that look like 1 horse)

Gestalt psychology: Closure

Closed contours are grouped together

Gestalt psychology: Common fate

Things which move together are grouped together

Why do Gestalt laws work

They reflect assumptions about the nature of real-world surfaces and objects.

Link the description of why Gestalt laws work with the 5 laws:

• Objects are made of cohesive, opaque, uniform materials

• When objects change or move, all their parts change or move

• Object contours tend to vary smoothly

• Proximity, similarity

• Common fate

• Good continuation

Three proposed routes to segmentation

1. Texture-based segmentation (based on FRF processing)

2. Motion- and depth- based segmentation

3. Symbolic segmentation

Contour integration: 1, 2, 3

1. Ending contour

2. Transverse contour T-junction

3. Generic contour (fold)

Why do concave elliptic patches never appear on the contour?

Their surface curves inward and never lines up with the viewer’s eye in a way that would make an edge visible.

How does symbolic integration work in the brain

(Such as Marr’s theory) connect basic features into coherent shapes, helping us recognize objects in the scene

Why is the figure on the right not ambiguous in comparison to the left figure in terms of shape recognition

There are only T-junctions instead of 4 way junctions

The Wollaston effect

Perceived gaze direction of a portrait depends not only on the position of the irises but also on the orientation of the head

The Mona Lisa effect

Perceived gaze direction of a portrait remains aimed at the observer despite changes of viewing angle

Object representation: Intrinsic factors

Define the character of a specific object (shape, surfaces, parts)

Object representation: Extrinsic factors

Relate to variation in viewing conditions (position, lighting)

View-independent theories try to (…) extrinsic variation and build a representation that captures the (…) of each object

Remove, intrinsic character

View-dependent theories (…) extrinsic effects into a (…) which represents the object

Combine, class of views

Object representation: Marr and Nishihara

3D generalized cones as part descriptors

Object representation: Biederman

Template matching of shapes/geons

Limitations of view-independent theories

• No possibility of fine discriminations

• Some object may have several, one or no possible decompositions (Theories have been difficult to implement in computer systems)

Object representation: Ullman

Recognition happens when the shape fits after transformation of stored view

Object representation: Edelman and Duvdevani-Bar

Multidimensional feature space representing the many dimensions along which shapes can vary

Limitations of view-dependent theories

Storage inefficiency and generalization difficulty