|
(a)Briefly describe the three stages of comutational vision: image
acquisition ,low level vision and high level vision .[3 marks ]
i.Image acquisition: the image is captured
and converted into digital form that can be processed by a computer.(1)
ii.Low-level vision: properties from the image representation, such
as edges, are extracted.(1)
iii.High-level vision:information extracted by low-level vision is
used for problem solution, such as navigation.(1)
(b)High and low level vision can be compared to each other in the
four areas of purpose , evaluation ,knowledge and processing mode
.For three of these areas,distinguish between high and low level vision.[3
marks ]
i.The purpose of low level vision is to recover
properties from the surrounding environment while the purpose of high
level vision is for solving complex tasks like navigation.(1)
ii.Low level vision is evaluated on the basis of validity while high
level vision is evaluated on usefulness.(1)
iii.Low level vision uses knowledge of the physical properties and
continuity of the object while high level vision uses knowledge acquired
about the object (like shape and transformation)(1)
iv.The processing mode of low level vision is parallel and bottom-up
while that of high level vision is goal directed and spatially focused.(1)
(Max 3 marks)
(c)Two common techniques of pattern recognition are template matching
and feature analysis.For each of the following applications,state
which of the two techniques is more appropriate,and justify your answers.
(i)Recognition of cheque numbers [3 marks ]
Template matching.The number of object to
be recognised is small, i.e.10 for the number of digits.(1)The size
and orientation is fixed allowing easy recognition.(1) The font of
the digits is well defined, requiring only a small number of templates.
(1)The position of the digits on the cheque is fixed, thus allowing
individual digits to be extracted easily.(1)(Max 3 marks)
(ii)Recognition of characters in magazine articles [3 marks ]
Feature analysis. In magazine articles,many
fonts of different sizes and
orientation may be used.(1)If template matching were to be used, a
large number of templates would have to be defined,and even then,
it will not be possible to define all possible variations.(1)On the
other hand, feature analysis requires the extraction of the features
of each character, which is almost always the same, regardless of
font face and ize.(1) Additional processing may be required forcharacters
of dfferent orientation.(1) More processing is required for feature
analysis then template recognition, but it would still be more efficient
than searching through a very large number of template .(1)(Max 3
marks)
(d)Describe how the letter Z can be recognised using feature extraction
and spatial relations.[3marks ]
Features that can be detected in Z are that
it consists of two horizontal lines and a line with a positive gradient
(sloping from lower left to upper right).(1) Spatial relation between
the three lines is that the two horizontal lines are located one on
top of the other. The bottom of the sloping line connect to the left
end of the lower horizontal line, and the top of the sloping line
connect to the right end of the upper horizontal
line.((2) for a complete description of the connection )
|