The way animals, especially humans view, see and are aware of the world that surrounds them is called perception. Processes in visual perception have a great importance on how an object is perceived. Light from the perceived object hits the retina from where it gets converted into meaningful information. Different theorists proposed a number of theories that try to explain how visual perception takes place. Anything from template and feature for pattern recognition theories, to computational and neuropsychological theories for object recognition which look at perception for the purpose of recognition have been proposed. In order to assess whether Marr's (1982) theory, together withÂ the supporting theories ofÂ Biederman (1987) and Riddoch & Humphreys (2001), provide both a valid andÂ a complete account of perception, one must look more specifically at these theories and how they explain perception. As well one must consider the strengths and limitations of these theories, what will lead to final conclusion on validity and compliance of the theories of perception.
Marr's computational theory of object recognition is one of the most influential theories of perception in Cognitive psychology, with no serious challenges and little derivatives (Costall, 1984). This theory was based on the development of a computational model that could explain the processes involved in object recognition. Processes that start at the point of light from an object hitting the retina of perceiver, all the way to "construction" of perceived object and its comparison with stored objects in a memory, what leads to object recognition. Once the object is identified, information about depth and distance is processed in the 2.5-D representation. At the end a full 3-D representation (which is view invariant) is formed what is compared with stored objects in memory (Marr, 1982; cited in Eysenck and Keane, 2001). Marr and Nishihara (1978; cited in Eysenck and Keane, 2001) devoted separate project to the matching of the 3-D representation with the information stored in memory. In which they pointed out that one first identifies concavities, which help to segregate different parts of the image what lead to identification of the main axis of the image. This enables the comparison with the stored memory in a viewpoint-invariant way. Therefore we can say that Marr viewed recognition as the aim of visual perception.
Biederman (1987) followed Marr's theory, especially the work done by Marr and Nishara (1978). Biederman developed his Recognition-by-Components theory, in which the emphasis is on the idea that objects are made of basic shapes, simple volumetric entities which he called geons (geometric ions), shapes like blobs, cylinder and edges. It is believed that around 35 geons are needed to explain the composition of most of the objects. These geons allow for better extraction of information, despite the occlusion and noise. Recognition of an object happens by analysis, identifying its main geons and the relationship between them, which then are matched with objects stored in memory. Another important aspect of Biadermans theory is the non-accidental properties of the 2 dimentional image of the geon seen by the observer. This holds that the visual input are the same as real, non - accidental, regularities in the environment, instead of being seen as accidental attributes of certain viewpoint or alignment of objects (Beiderman, 1987). For Biederman (1987) believed that perception occurs in following stages. First stage is the edge extraction from the visual input. Second stage is detection of non-accidental properties and parsing of the image at concavities. Third stage is determination of components, followed by fourth stage, the matching of components with stored objects representation. The last stage is when the object is finally identified. As well as Marr, Biederman sow recognition as important part in visual perception.
Riddoch and Humphereys' (2001) theory of Perception and Object recognition as well as Biederman's theory, is derived from Marr's theory. The aim of this model is to fit with the data collected from patients with neuropsychological impairments, which are cause of impairments in visual perception and object recognition (Eysenck and Keane, 2001). In this theory visual object agnosia is looked into in more detail, since it helps to understands perception. In this theory each stage depends on the output of the previous one. The stages they identify are very similar to Marr's theory. In this model, firstly we have four modules by which analysis of an object are carried out. This models are motion features, colour features, form features and depth features, which reflect what is known at visual level about object perceived (Eysenck and Keane, 2001,p.87). In the next two stages, edge grouping collibearity and feature binding into shape, the concentration is on form analysis. At this point the features of the object observed are made into larger structures so the figure can be better recognised. If view of object is recognised then it is send directly to Structural Description System. If the object is not recognised or of unusual shape then it is "normalised" what is called View normalisation from where is forwarded to Structural Description System. At this point the percept is matched with images stored in our memory. From this stage we move to Sematic system in which structural information about the object represented are stored. Finally, the name of the object is identified (Eysenck and Keane, 2001,p.87). The theory tries to explain the normal and abnormal functions in visual perception and object recognition, which are well supported by the neuropsychological data.
The three models discussed above describe very important and relevant ideas about perception. However, it is important to look at the supporting and conflicting evidence for all three arguments in order to understand the validity of these theories. To start with, Marr's and Biederman's theories in short, tried to address and to accept the complex details of object recognition, which has so far been oversimplified. One of the tasks that support their hypothesis is where participants had to decide if a line drawing matched a second presented drawing, in which the line seemed to be shorter, what had an influence on a performance (Lawson and Humphreys, 1996; cited in Eysenck and Keane, 2001). This task supports the hypothesis of detecting major axis in object recognition. Another support comes from Hummel and Biederman's (1992; cited in Eysenck and Keane, 2001) connectionist model, which has proved to be very sufficient model for explaining the binding problem. More support comes from Cooper and Biederman (1993; cited in Eysenck and Keane, 2001) when it comes to explaining the importance of geon detection. Further evidence comes from experiment done on monkeys, where Vogels, Biederman, Bar and Lorincz (2001; cited in Eysenck and Keane, 2001) suggested that existence of geon-sensitive neurons. Nevertheless, cautious has to be applied when it comes to taking into account these pieces of evidence. As one may have noticed, Biedermans in a number of these researches played a role what may have led to influence of results.
When it comes to supporting Riddoch and Humphereys' (2001) theory of Perception and Object recognition, (Fery and Morais 2003) reported a patient who was very good at tasks that involved recognising and matching shapes, but was able to identify only 16% of these objects. Another support comes from study done by Hillis and Caramazza (1991) on patient with category specific deficits. This provides the ability to recognise living things while having the inability to recognise non-living things. Further support comes from study done by Humphreys and Riddoch (1987) on patient HJA which suggests a deficiency in feature binding and shape segmentation. Miller et.al (1991) looked into damage to the "grouping by colinerarity" which results support Riddoch and Humphereys theory. Finally, neuropsychological patients with difficulties to recognise objects from unusual views indicate the ability to rotate a three dimensional representation is a key feature of object recognition (Warrington and Taylor; 1978; cited in Eysenck and Keane, 2001). This theory is strongly supported with neuropsychological data, which will help it improve and define in the future.
In spite of the evidence supporting all these three theories, there are several instances of disconfirming evidence. Starting with Marr's and Biaderman's view; Joseph and Proffitt (1996; cited in Eysenck and Keane, 2001) have pointed out the advantage of colour information on object recognition that is present in many studies. This does not support Biederman's idea that edge information is important for object recognition, rather than surface information, like colour. One of the key evidence against Biederman and Marr's theory is the fact that many things in the world do not have a geon structure. e.g. sky, sea. Marr's theory like Biederman's is not very suited when it comes to recognising object within category. e.g. faces. Support for Marr's and Biederman's theory is very inconsistent. Moreover, these theories do not consider the importance of environment on perception of an object. Palmer (1975; cited in Eysenk and Keane, 2001) proposed how important is context when it comes to influencing object recognition. Recognition is easier if object is harmonious with the environment, average if there is no context, and much more difficult if there is no consistency. When it comes to Riddoch and Humphreys' theory, even though it certainly explains well visual agnosia, there are still some limitations. Firstly, the theory should be seen as a framework rather than complete theory (Eysenck and Keane, 2001), since the stages using output from previous stage is not clearly explained. Secondly, the assumption is that recognition happens trough bottom-up way. However, one should not dismiss the fact that it is likely that there are some top-down processes during object recognition (Eysenck and Keane, 2001).
In conclusion, different theories tried to explain perception. They all added great value to our understanding of perception and recognition. However, there is still considerable number of evidence that weakens these theories. Where Marr and Biaderman defended object recognition, others pointed out the importance of the environment on the perception. One of the major consideration to be taken into account is the evidence provided in large has been produced by research groups that included the theorists, so they may be considerably biased in designing the experiments, or in their interpretation. ...........................