L5 object recognition

Object Recognition

Course Information

  • Course: PSYC 236

  • Topic: Cognition and Perception

  • Presenter: Özge Karakale

Key Concepts in Perception

Perceptual Segregation
  • Definition: The process of separating visual input into distinct objects.

  • Importance: It is believed to occur prior to the recognition of objects.

Gestalt Psychology
  • Overview: A perspective in psychology that emphasizes the human ability to perceive patterns and wholes in sensory data.

  • Law of Prägnanz: Proposes that perceptual organization will always be as regular, simple, and symmetrical as possible.

Gestalt Laws of Perceptual Organization
  1. Proximity: Objects that are close together are perceived as belonging to the same group.

  2. Similarity: Objects that are similar are grouped together.

  3. Continuity: Lines are perceived to follow the smoothest path.

  4. Closure: The mind tends to perceive incomplete shapes as complete.

Figure-Ground Segmentation
  • Definition: The process by which the visual system distinguishes between the object (figure) and the background (ground).

  • Influencing Factors:   - Attributes such as color and luminance.   - Characteristics of figures:     - Convex shape (curving outward).     - Dimensions (small size).     - Surrounded by other shapes.     - Symmetrical layout.   - Figure vs. Ground:     - Figure: The object of interest.     - Ground: The background.

Real-World Applications of Gestalt Principles

Adjacent Contour Segments
  • New Principles:   - Adjacent segments of any contour tend to exhibit similar orientations.   - Segments that are farther apart have slightly differing orientations.   - Knowledge of real object properties affects contour decision-making.

Uniform Connectedness
  • Definition: Any connected region with uniform visual properties (color, texture, etc.) is often perceived as a single perceptual unit.

  • Implication: Can overpower traditional Gestalt grouping laws such as proximity and similarity.

The Accentuation Principle
  • Definition: Lines of dissimilarity play a significant role in perceptual organization.

  • Reference: Pinna et al. (2018) describe this as a fundamental principle for figure-ground segregation.

Influences on Object Recognition

Nature vs. Nurture
  • The assumption in Gestaltism was that figure-ground segregation is innate. However, memory (object familiarity) plays a crucial role in recognition.

  • Study Reference: Barense et al. (2012)   - Healthy controls could identify familiar objects as figures, while amnesiac patients struggled.

Advantages and Limitations of Gestalt Theory

Advantages
  • Focus on essential issues in perceptual organization.

  • Many organizational principles have remained relevant and valid.

  • The Law of Prägnanz proves to be highly beneficial.

Limitations
  • Underestimation of the role of knowledge in perception.

  • Primarily descriptive; lacks explanatory power.

  • Based predominantly on two-dimensional artistic representations.

  • Does not adequately address complex interactions among grouping laws.

  • Difficult to empirically test the Gestalt approach.

Processing Flow in Vision

Spatial Frequency Processing
  • Description: A method categorizing visual data into low spatial frequencies (LSF) and high spatial frequencies (HSF).

  • Concept: Coarse-to-fine processing allows for initial global layout followed by detailed analysis.

Retinal Origin and Cortical Mapping

Spatial Frequency

Retinal Origin

Cortical Mapping

Function

Low (LSF)

Peripheral

Anterior V1

Global layout, fast scene gist

High (HSF)

Foveal

Posterior V1

Fine details, object recognition

Major Theories of Object Recognition

Cognitive Theories
  1. Marr’s Computational Approach (1982)    - Staged processing of visual information:      - Primal Sketch ➜ 2.5D Sketch ➜ 3D Model.

  2. Biederman’s Recognition-by-Components (1987)    - Object recognition achieved through geometric icons known as geons and emphasizes viewpoint invariance.

Neuroscientific Theories
  1. Ventral vs. Dorsal Pathways    - Ventral pathway: Responsible for object identification (What).    - Dorsal pathway: Concerned with spatial awareness and actions (Where/How).

  2. Global vs. Local Processing    - Focuses on how the brain processes global shapes versus local details, inspired by Navon (1977).

  3. Predictive Coding / Bayesian Brain    - The brain processes perception through inferences based on a combination of sensory evidence and prior experiences.

Marr’s Computational Approach

  • Primal Sketch: A two-dimensional depiction of light-intensity variations, which encapsulates edges, contours, and blobs.

  • 2.5D Sketch: Captures depth and surface orientation through shading and texture.

  • 3D Model: Represents the object’s shape in three dimensions, allowing for viewpoint independence.

Biederman’s Recognition-by-Components Theory

  • Objects are composed of combinations of geons, which are elementary geometric shapes.

  • Recognizing objects occurs through the identification of these components based on non-accidental properties across different viewpoints.

Strengths and Limitations of Theories

Biederman’s Recognition-by-Components
Strengths
  • Strong evidence supporting the importance of geons in recognition tasks.

  • Significant role of concavities and edges in identification.

  • Non-accidental properties aid in recognizing objects from various angles.

Limitations
  • Insufficient focus on top-down influences like context and knowledge.

  • Inability to account for nuanced within-category differentiations (e.g., specific dog breeds).

  • Acknowledgment that viewpoint-dependent recognition is common.

  • Certain objects lack invariant geons yet remain easily recognizable (e.g., clouds).

Zeki’s Functional Specialization Theory
  • Visual Processing Areas:   - V1 & V2: Basic visual processing.   - V3 & V3A: Form perception, particularly for moving stimuli.   - V4: Responsible for color processing and line orientation.   - V5 (MT): Concerned with motion perception.

Building on Zeki's Insights: Ungerleider & Mishkin (1982)
  • Proposed the distinction of two primary processing streams:   - Ventral Stream: Involved in object recognition.   - Dorsal Stream: Involved in spatial processing.

Goodale & Milner (1992, 1995)
  • A model integrating specialized modules focused on visual functional goals:   - Ventral Stream: Vision for recognition and identification.   - Dorsal Stream: Vision for action and interaction in physical space.

Perception-Action Model

Characteristics of the Streams
Ventral Stream
  • Characterized as the “What” pathway.

  • Involves processing in the occipital lobe extending into the temporal lobe.

  • Functions in form and color processing, enabling recognition linking with memories.

Dorsal Stream
  • Characterized as the “Where/How” pathway.

  • Extends from the occipital lobe into the parietal lobe.

  • Primarily involved in motion processing for action guidance in real-time situations.

Selectivity of Visual Cortex Cells

  • A study indicated the response rates of specific areas to features such as orientation, direction, disparity, and color.

  • Evidence demonstrated that visual perception occurs in highly specialized areas of the brain.

Functional Specialization: Evaluation
Strengths
  • Enhanced understanding of visual processing roles and connections.

  • Highlighted varied functionalities across different brain regions.

Limitations
  • Demonstrates that the complexity of visual brain operations transcends original theories proposed.

  • Observations of patients with visual impairments often show mixed capabilities, indicating overlap.

  • The integration of information, also known as the binding problem, poses questions about unified sensory perception.

Interactive-Iiterative Framework of Object Recognition

Insight into Top-Down vs. Bottom-Up Processes
  • Illustrates how top-down cognitive influences impact attention, which then affects bottom-up visual processing.

  • Notably, top-down processes become more significant in interpreting degraded or ambiguous stimuli.

Bayesian Perception

Conceptual Overview
  • The brain acts as a prediction machine that infers the best guess of reality based on sensory evidence and past experiences.

  • Formula:   - Perception = Inference   - Incorporates:     - Sensory Evidence: Variable and often noisy.     - Priors: Past experiences and contextual expectations.     - Outcome: A refined estimation of reality.

Applications of Bayesian Perception
  • Example: Assessing whether an object on a messy desk is a cup:   - Prior: Knowledge suggests a high probability of encountering a cup.   - Likelihood: Observing a rounded shape with a handle, while also considering other possible interpretations.   - Posterior: Integrating prior expectations with sensory information yields the conclusion, “That is a cup.”   - Prediction Error: The difference between anticipated and perceived observations.   - Uncertainty and Noise: Factors that influence the clarity of recognition.

Conclusion

  • The interaction between perceptual processes and cognitive frameworks shapes how we visually recognize and interpret objects in our environment. Insights from various theories, such as Marr's and Biederman's, aid in crystallizing our understanding of the complexities inherent in object recognition. Top-down processing plays a vital role, especially under ambiguous visual circumstances, signaling the need for both perceptual theory and empirical evidence in cognitive research.