PSGY1010 Cognitive Psychology 1: Perception PDF

PSGY1010 Cognitive Psychology 1 Perception III: Depth and Visual Scene Analysis Dr Chung Kai Li [email protected] Tell my what you see… 2 Perceiving machines 3  Why is it difficult to design a perceiving machine? ▪ Visual input...

PSGY1010 Cognitive Psychology 1 Perception III: Depth and Visual Scene Analysis Dr Chung Kai Li [email protected] Tell my what you see… 2 Perceiving machines 3  Why is it difficult to design a perceiving machine? ▪ Visual input provides ambiguous information about the 3D structure of the world The same 2D retinal image could be produced by an infinite This ambiguity is exploited by number of 3D objects anamorphic street artists 4  Why is it difficult to design a perceiving machine? ▪ Image complexity makes it increasingly tricky for computers to organise the visual scene into distinct objects Image Output of edge detection algorithm 5  Today’s lecture Learning objectives: ▪ Understand how we extract information about 3-dimensional space from 2-dimensional retinal images ▪ Compare the Structuralist and Gestalt approaches to perceptual organisation and describe the principles of object grouping 6  How do we perceive depth? ▪ Humans make use of a variety of sources of image information to infer depth in a visual scene ▪ Monocular cues (work with one eye) ▪ relative height ▪ relative size ▪ occlusion ▪ linear perspective ▪ texture gradient ▪ motion parallax ▪ shadows ▪ shading ▪ Binocular cues (require both eyes) ▪ disparity 7  Relative height and size ▪ Objects that are below the horizon and have their bases higher are typically perceived as being more distant ▪ If two objects are of equal physical size the more distant one will take up less of your field of view ▪ We need prior knowledge about the relative sizes of objects when judging distance 8  Occlusion ▪ Closer objects will occlude further away ones Relative height, relative size and occlusion cues all contribute to our perception of depth in this image 9  Linear perspective and texture gradient ▪ Parallel lines extending away from ▪ Texture elements get smaller observer converge in the distance and more dense with distance ▪ Foreshortening (circles become ovals) occurs when the surface is tilted away  Motion parallax ▪ As we move, more distant objects will glide past us more slowly than nearer objects 11 Shadows and shading Brightness of a surface depends on its orientation with respect to the light source The addition of cast shadows to the Note – when the image above is viewed upside down, bottom video creates a strong many people perceive the footprints as flipping from perception of depth concave to convex (light-from-above assumption) 12  Binocular disparity Stereoscopic vision ▪ Our two eyes receive a slightly different image of the world Disparity ▪ This creates a differences in image location of an object seen by left and right eyes ▪ The size of the disparity depends on an object’s depth Try lining up your index fingers with one another, then closing each eye in turn 13  Binocular disparity Horopter ▪ Set of points in space that project to corresponding positions in the two retinas (i.e., zero disparity) ▪ Includes the fixation point When the lifeguard looks at Frieda, the images of Frieda, Susan and Harry fall on corresponding positions on the lifeguard’s retinas (if one retina could be slid on top of the other, the points would overlap). 14  Binocular disparity Uncrossed disparity ▪ Objects closer than the horopter have crossed disparities ▪ (You would have to cross your eyes to fixate on it) ▪ The image lies further to the right from the right eye’s viewpoint than from the left eye’s viewpoint ▪ Objects further than the horopter have Crossed uncrossed disparities disparity ▪ (You would have to uncross your eyes to fixate on it) ▪ The image lies further to the left from the right eye’s perspective 15  Summary Learning objective: Understand how we extract information about 3D space from 2D retinal image ▪ A variety of depth cues are contained in each monocular image, including relative object height and size, occlusion, linear perspective, texture gradients, motion parallax, shadows and shading ▪ Perception of depth relies on a combination of bottom up (extraction of the cues) and top-down (e.g. prior knowledge of object size, lighting direction) processes ▪ Depth information is also extracted from binocular disparities, via a process known as stereopsis 16  How do we perceive objects? Structuralism ▪ Approach pioneered by Wilhelm Wundt that was popular in mid to late 19th century Wundt (1832-1920) ▪ Proposed that perceptions are simply the sum of ‘atoms’ of sensation 17  How do we perceive objects? The Gestalt school ▪ Reaction against structuralism led by three central figures working at Frankfurt University: Wertheimer, Köhler & Koffka ▪ Argued that the whole form or configuration is greater than the sum of its parts Wertheimer Köhler Koffka 18 (1880-1943) (1887-1967) (1886-1941)  How do we perceive objects? Illusory contours ▪ some images evoke the perception of edges in locations where there is no change in luminance or colour ▪ difficult to explain via the structuralist approach Do you perceive an upwards pointing arrow? Do you perceive a floating cube? 19  The Gestalt principles of perceptual organisation ▪ Having rejected structuralism, the Gestalt psychologists proposed a number of principles by which elements in an image are grouped to created larger objects ▪ Principles are all manifestations of the Law of Prägnanz (‘good figure’) ▪ “Of several geometrically possible organisations, that one will actually occur which possesses the best, simplest and most stable shape” (Koffka, 1935) 20  Proximity ▪ Things that are close together group together ▪ A perceived as rows as horizontal spacing is smaller ▪ B perceived as columns because vertical spacing is smaller ▪ C is ambiguous as dots are equally spaced A B C 21  Similarity ▪ Things that are similar group together ▪ Similarity could be in terms of any basic characteristic, such as shape (A) or orientation (B) ▪ The perception of (C) as columns occurs even though the proximity information suggests rows – similarity may override proximity A B C 22 Similarity versus proximity 23  Common fate ▪ Things that move together group together Strong grouping from common fate & proximity Motion can be a powerful way of segmenting objects from the background Weaker grouping common fate alone 24  Good continuation ▪ Group elements to form smoothly continuing lines rather than abrupt or sharp angles ▪ Helps preserve grouping of occluded objects 25  Closure ▪ Group elements to form complete figures, even if incomplete Grouping by closure requires top-down knowledge of forms 26  Symmetry ▪ Elements more likely to be formed into groups that are balanced or symmetrical 27  Assessment of Gestalt principles ▪ Seem correct about many things – perceptual objects are not simply the sum of their parts ▪ Gestalt principles generally hold across wide range of images ▪ However, some of the principles seem rather vague and imprecise ▪ not always clear what is meant by a ‘good’ or ‘simple’ shape ▪ No coherent workable account of the underlying neural mechanisms ▪ Köhler proposed an electrical field theory, in which lines of flow are created in the brain which match the structure of perception ▪ No empirical evidence for this! 28  Summary Learning objective: Compare the Structuralist and Gestalt approaches perceptual organisation and describe the principles of object grouping ▪ According to structuralist theorists, perceptual objects are formed by grouping a number of primary sensations ▪ Gestalt psychologists argued that objects are stable, organised wholes ▪ They proposed that features are grouped according to to key principles including ▪ proximity ▪ similarity ▪ common fate ▪ good continuation ▪ closure ▪ symmetry 29 Thank you Any questions?