Dec1999 CO230 : Cognitive Science (Question 4)Suggested Solutions

(a)

Compare the high level and low level vision processes with respect to:

(i)the type of knowledge used in each process [1 mark ]

The low level vision process uses knowledge about the general physical properties of objects,while high level vision uses knowledge already acquired about specific objects.(1)

(ii)the type of processing performed in each process [1 mark ]

Low level vision makes use of a parallel mode of processing while processing in high level vision may be sequential in nature.(1)

(iii)the goals of each process [1 mark ]

The goal of low level vision is to recover physical properties from the environment while the goal of high level vision is to make use of the information obtained to solve a task.(1)

[3]

(b)

Template matching and feature analysis are two techniques commonly used for performing pattern recognition.

(i)Describe the template matching technique.[3 marks ]

Variations of images of an object,known as a templates,are stored in the computer (1). Incoming information is compared with the templates (1)until a good fit between the incoming information and the stored representation is found (1).

(ii)Describe the feature analysis technique.[3 marks ]

The object is represented by the features that comprise it,and their relative position (1).The distinctive characteristics of the input image are extracted (1). The presence and absence of features are counted and compared to the existing table of features (1).

(iii)Banks can use template matching for recognising cheque numbers.Describe the properties of this application that make it suitable for template matching pattern recognition.[3 marks ]

Accept other correct answers -maximum 3 marks
The character set that is used in cheques is standardised using a standard font set e.g.OCR-A (1).The location of each character that must be recognised is found in a pre-defined position (1).The size of the font used is fixed (1).The orientation of the characters to be recognised are fixed (1).The total number of characters to be recognised is fixed (1).

(iv)Facial recognition systems for security purposes generally do not use template matching.Describe the properties of this application that make it unsuitable for template matching pattern recognition.[3 marks]

Accept other correct answers -maximum 3 marks
The features of the face may vary (e.g.due to injury,hair style etc)(1).The orientation of the face with regards to the camera will be different every time the image is captured (1).The distance between the camera and the face will not be exactly the same (resulting in different size of the face with respect to the image) (1).This makes it difficult to produce a set of templates that can match the person ’s image at all times.

[12]