Q2
2-1.
i) In image analysis, which features, colour, shape, texture, and spatial relationship play important roles in recognising the content of image a), b), and c). Please explain your analysis. (6 marks)
ii) What additional information do you think may be useful to interpret the meaning of these sample images if the above feature content (colour, shape, texture etc) is not sufficient enough for the computer to understand the images? (4 marks)
2-1-a) Body tissue under the microscope
(can be rotated at any angle)
2-1- b) a set of logos.
2-1-c) a house by the pond
Solution:
i)
The answer to this question can be flexible as long the students give reasonable analysis about which features are important in the sample images.
A)the texture in different region, the shape of different objects and colour are possible the most important features in image a). Since it can be rotated at any angle, and there is not specific spatial constraint between objects, the spatial relationship may not so important.
B)Shape should be the prime feature in such series of images as the shape of the object decides what the object is. The colour may play the role to identify the contour of the objects.
C)Shape, spatial relationship, colour play roles in this image. Since the texture in each region is not so obvious, texture may be the least important feature in this image.
ii)
Domain knowledge will be helpful to further understand the images. In high-level applications, domain knowledge is always modelled and applied to assist an intelligent analysis. The representation of knowledge can be in many different forms (frames, rules, semantic network/relational structure etc) that may serve for bottom-up or top-down or hybrid analysis. The key in modelling knowledge is to enable the system understand the associations between the extracted primitives and their high-level meanings.