PS2111 CDL Lecture 1.pptx

PS2111 Information processing and cognition Carlo De Lillo George Davies Centre, Room 3.12 Email: [email protected] Outline How we encode and represent information Interplay of functions in cognition 5 lectures + revision Pattern and object recognition Visual guidance of action Short-term and working me...

PS2111 Information processing and cognition Carlo De Lillo George Davies Centre, Room 3.12 Email: [email protected] Outline How we encode and represent information Interplay of functions in cognition 5 lectures + revision Pattern and object recognition Visual guidance of action Short-term and working memory Mental imagery Visual systems for perception and for action Stimulus equivalence problem The same object can produce an infinite number of retinal images Different object can produce the same retinal image How does the cognitive system decide which retinal images are of the same object? How does the cognitive system decide which parts of an object belong together? A A A AA A A AA A Pattern recognition Simple recognition mechanisms Tinbergen 1951 Aggressive behaviour of male stickleback Red belly patch triggers response Detailed model of rival fish is ineffective Recognition of complex shape reduced to one key feature More sophisticated visual recognition processes require matching complex stimulus configurations with internal representations pattern recognition of 2D shapes is a simplified problem of object recognition applied problem motivated by machine vision main approaches Template theories Feature theories Structural theories Template theories templates of patterns in memory stimulus patterns are compared with templates if they match the template, then the stimulus is recognised Problems with template theories matching would fail even with minor differences between stimulus and template A stimulus find major axis rotate A Standardisation processes are required e.g. A scale size A scale line weight A template A even after standardisation mismatches are likely to occur stimulus templates matching A R A R More serious problems of templates theories How to decide which features to modify to enable matching E.g. should these patterns be modified to match a Q or and O? Experimental evidence against template matching (Sutherland and Williams 1969; Sutherland 1973) Rats trained to discriminate between a+ and b Transfer test with c and d Rats trained on a+ select c, even if d is d overlaps better with a Rats represent the structure (regular irregular) of a pattern rather than a matching template Feature theories Influenced by Hubel and Wiesel’s discovery of feature detector cells in the primary visual cortex Selfridge’s (1959) “pandemonium” exemplifies many different theoretical models using this approach Problems with feature detection recognition based on lists of features no information about different instances of a patterns Inability to distinguish patterns with same features in different configurations Image from: https://www.flickr.com/photos/appelogen/5179264587/in/photostream/ Warhol Warhol Warhol Warhol Structural theories more an approach than a theory roots in Gestalt psychology set of representations propositional (statements with objects and predicates) explicit description of relationships can be abstract and robust enough to represent patterns and2D and 3D e.g. letter T vertical line, horizontal line (supports, bisects) Object recognition We face more complex problems than with pattern recognition with 3D objects in real world scenes Which memory representations can support reliable recognition? View independent theories Marr’s Biederman’s View dependent theories Marr’s theory of object recognition Complete computational theory Mostly bottom-up Modular theory Independent processes producing different representations Gey level description Primal sketch 2½D 3D Key notions in Marr’s approach Generalised cones Key processes in Marr’s approach mage segmentation and derivation of axes Why axes are important What is this? Stages in Marr’s theory of object Imagerecognition segmentation Derivation of major axes Determining generalised cones Matching with 3D templates in memor Biederman’s recognition by components theory Library of 36 primitive 3D shapes “geons” Defined by nonaccidental properties Collinearity Symmetry Etc. Structural descriptions Combining geons an infinite number of objects Biederman’s (1987) recognition-by-components theory Edge extraction Detection of non-accidental properties Parsing of regions of concavity Determination of components Matching of components to object representation Objects structural models with details of spatial arrangement, size etc. of geons Evidence supporting recognition by component Regions of most concavity or vertices (b) are more important for identification of objects (a) than mid-regions of segments with less curvature (c) (Biederman, 1985, 1987) The most important regions are those that define relationships between geons Have these two objects the same name? Have these two objects the same name? Evidence supporting recognition by Metric change relative component to index of geon change Geon Original change Cooper and Biederman, 1993 Participants to judge whether two objects presented in rapid succession have the same name (e.g. “hat”) Longer RT and lower accuracy with geon change than metric change Geon change more - 50% Same + 50% View-point dependent and view-point invariant approaches to object recognition View invariant View dependent 3D representation of objects Predicts equal ease of recognition at most viewpoints Object represented as collection of views Predicts easier recognition when objects seen at those orientations View-point invariant versus view point dependent approaches Evidence of view-point invariant representations (Biederman & Gerhardstein 1993) – Participant asked to name images of familiar objects as quickly as possible – Presentation of to be named object positively primed by objects presented in different orientations Evidence of view-point dependent representations – unfamiliar objects such as “Greebles” Tarr & Bulthoff (1995); Gauthier & Tarr (2002) – participants familiarised with Greebles in given orientations are slower in identifying them when presented in different orientations Object categorisation and identification Review by Milivojevic (2012) Categorisation Concept (e.g. is this a dog?) Not slowed down by changes in orientation/viewpoint Identification Exemplar (e.g. is this “Billy”?) Slowed down by changes in orientation/viewpoint changes from canonical view Only representations of for recognition of object categories may be viewpoint invariant Summary Stimulus equivalence problem Approaches to pattern recognition Marr’s approach Biederman’s recognition-bycomponents View-point dependent/invariant approaches Reading Essential Bruce, Vicki; Georgeson, Mark A ; Green, Patrick R 2003 Visual Perception: Physiology, Psychology and Ecology. Chapter 9: Object recognition. Additional Reed, Stephen, K. Cognition: Theories and applications, 9th edition 2013. Chapter 2: Pattern recognition Eysenck, Michael W ; Keane, Mark T 2010 Cognitive Psychology: A Student's Handbook, 6th Edition. Chapter 3.