Medin, Ross and Markman (2001, p. 97) state that the key principle underlying feature comparison theories is ‘that all objects are composed of separable, distinct parts referred to as features’. The hypothesis that we analyse and recognise objects with regard to distinctive features is supported by Neisser (1964, in Eysenck & Keane, 2000). He found that when asked to identify a target letter (Z) embedded within distracters that shares the same features, for example X and E, it took people longer to locate the letter Z than when it was embedded within letters that did not share the same features, for example C and O. This experiment shows that distinctive features must play a role in pattern recognition; if they did not it would have taken the same amount of time to find the target letter within both groups of letters. An advantage of feature models is that they also analyse the extent to which a feature is present, not just whether a feature is present or not. Therefore, even if a feature is not perfect the pattern can still be recognised. This helps explain how we are able to read different handwriting.
The first real feature analysis model was proposed by Selfridge (1959, in Payne and Wenger, 1998) and is one of the most influential models within feature theories. The pandemonium model essentially involves a hierarchical system of demons that perform various information processing tasks that eventually lead to the recognition of an image. The pandemonium model involves several simultaneous stages compared to template models, which only involve two stages and looks at more detail when analysing what the particular pattern/object is. It also has the advantage over template models in that with a limited set of feature detectors it can recognize a potentially infinite number of objects and it will also recognize letters in spite of changes in size, orientation, and other distortions. The pandemonium model also has advantages over earlier theories as it becomes more biologically plausible if one assumes that the demons function rather like neurons.
Further support of biological plausibility in feature models was provided by Hubel & Wiesel (1962, in Payne and Wenger, 1998) through their studies of individual cells in the retina and visual cortex. They measured the responses of the individual neurons in the visual cortex to stimuli projected onto the retina of a cat. In simple terms what they found was that there are specialised cells for feature detection in the visual cortex, with certain cells responding maximally to certain stimuli, for example straight-line stimuli in a particular orientation (Eysenck & Keane, 2000). Their research supports feature models as it demonstrated that the visual system builds an image from simple stimuli into more complex representations (Goldstein, 2001).
Although the pandemonium model and further findings by Hubel and Wiesel provide an idea of how features are detected and matched against mental representations in our memory, they do not explain how features are combined and recognised thereafter as actual objects in the environment. For this to occur top-down processing i.e. the idea that information and knowledge from higher cognitive processes guides lower level processing, would be required. Therefore a major limitation with feature models is that their emphasis is on bottom-up processing; i.e. that all processing develops from a detailed analysis of a pattern or object to a global analysis. Research has shown that global processing can often precede local processing and this is supported by Navon (1977, in Eysenck & Keane, 2000). He found that when a large letter (global level) was made up of smaller letter’s (local level) the time taken to respond to the small letters was slower when the large letter was different from the small letters. He also found that the response to the large letter was unaffected by the small letters. This suggests that some top-down processing must occur in pattern/object recognition.
Knowledge about objects has been found to be important in pattern/object recognition. Context and individuals past experiences and expectations can affect our perceptions and feature models do not account for this. The effects of context are shown with these two words, . The central letters in each word are exactly the same, yet we see one as an H and the other as an A, based on their context. We do this because we know, based on our vocabulary, that the first word must be THE and not TAE. This shows that knowledge and expectations can direct our perceptions, and explains how we can recognise objects even when some of their major features cannot be seen. This is emphasised by Biederman in his recognition-by-components theory. Biederman, Ju & Clapper (1987 in Eysenck & Keane, 2000) found that even when some of their components were missing, complex objects were still identified. Biederman’s theory also improves on feature theories in that it includes spatial descriptions. Feature models are limited as they do not consider the spatial relationships between features and as the same features can produce different patterns, in order to identify an image, in addition to knowing the particular features of it we need to know how these features are presented and connected (Eysenck & Keane, 2000).
In conclusion, although feature comparison models do a satisfactory job in explaining how we analyse images for features and match them with images stored in memory, they do not explain how features are combined and recognised thereafter as actual objects in the environment. Although supported by both behavioural and neurological evidence, feature models are limited as they do not account for top-down processes, and at best address only part of the process of pattern/object recognition.
1,085 words.
References
Eysenck, M. W., & Keane, M. T. (2000). Cognitive Psychology: A
student’s handbook (4th ed). Hove: Lawrence Erlbaum.
Eysenck, M. W. (1993). Principles of cognitive psychology. Hove:
Lawrence Erlbaum.
Goldstein, B. (2001). Sensation and Perception, 6th ed. London:
Wadsworth.
Medin, D. L., Ross, B. H., and Markman, A. B. (2001). Cognitive
Psychology (3rd ed). Orlando: Harcourt Brace.
Payne, D. G., & Wenger, M. J. (1998). Cognitive psychology. Boston:
Houghton Mifflin.