The subject matter of this assignment comes from two broad topics: image understanding and knowledge representation. The relevant reading for the image understanding part is the "Image-Analysis.pdf" file linked from the Textbooks page of our course website. The relevant reading for knowledge representation is from the chapter on knowledge representation in Part 2 of "The Elements of Artificial Intelligence Using Python", pages 99-116 (also linked from the Textbooks page).

You should turn in a plain ASCII text file named A2PartI.txt for Part I. (Updated on Oct. 14:) For Part III, turn in a Python file (similar to the MyLinneus.py file that you turned in for Lab 2). Parts II should be turned in as a document in Word (.doc or .docx) or Open Office (.odt) or PDF, with the name A2PartII.doc (or one of the other extensions). This document will contain a combination of textual answers, and embedded images.
 

1. IA 7. (about polar lines)
2. IA 8. (also about polar lines)
3. IA 11. (about connected components)
4. IA 15a. (about erosion)
5. IA 15b. (about dilation)
6. Design a structuring element for the hit-or-miss transform that fits into a 3-rows by 4-columns array that will detect the right endpoints of horizontal line segments.
7. KR 3. (about ISA hierarchies)
8. KR 5. (about partial orders). You can render the diagram using character graphics (be sure to use a monspaced font such as Courier or Courier New), or you can create (see Part II) and include a PNG image near the beginning of your document for Parts II and III with the label "Diagram for Part I problem 8".

1. (5 points) Apply Roberts' edge detection operator to the stamps and coins image (available from PixelMath in INFACT as "StampsAndCoinsReduced.jpg"). Save your results as Edges.jpg. Show the image in your document and answer the question: "To what extent have the edges of the stamps and coins been identified?"
2. (15 points) Use the Hough transform (the code is available in INFACT from the PixelMath Python File menu, "Load From Server" option) to find the lines that border the stamps in the image. Modify the code as necessary, and include your modified version of the code in your report. Show the Hough Transform (parameter space) image itself in your report. Then backproject the lines onto the original image. Save this as StampLines.jpg and include it in your report.

   What value did you use as a cutoff for deciding whether a peak in the Hough array was significant enough to consider as a stamp border line?

3. Optional, for extra credit (5 points): Modify an implementation of the Hough Transform so that pixels only vote for lines in a narrow band of directions. To determine the dominant direction of an edge, you should use the Sobel gradients, rather than the Roberts Cross operator. Then get the angle of the gradient using either the Python math.atan2 function or the PixelMath calculator function angle(dx,dy). The latter works only as part of a formula within the PixelMath calculator.
4. Optional, for extra credit (5 points). Modify your Hough Transform code so that it detects circles rather than lines. You may wish to set it up so that it takes the radius as a top-level argument. Your code can then call the Hought Transform routine several times to find circles of different radii. Apply your circle detection code to the StampsAndCoinsReduced.jpg image.
5. Optional, for extra credit (5 points). Using your circle detection code as a starting point, add functionality to output a list of hypotheses about how much money there is in the photo and an indication of the weight on that hypothesis. For example, one hypothesis might be that there is $1.43 showing in the image and another that there is $1.52. These two might differ because it is difficult to tell whether a particular circle represents a penny or a dime, which are, by-the-way, almost the same size and shape. Because the color of the coin is slightly brown, a higher weight may be put on the $1.43 hypothesis.

   After adjusting your system to get the right answer on the first image, try it on another coin image (Coins-abutting-reduced.jpg), and show the results.

   Include your code and its output in your report.

1. cycle detection (5 points). Have Linneus detect possible cycles in the database, so it can have the following type of dialog.

   > A a being is a creature.
   I understand.
   > A creature is an animal.
   I understand.
   > An animal is a being.
   Yes, but a being is an animal, because a being is a creature and a creature is an animal.
   I'm not going to remember that an animal is a being unless you insist.
   > OK
   I don't understand.
   

2. cycle processing (15 points). Have Linneus handle cycles by making one element of a cycle the representative in the ISA hierarchy and making others into synonyms without separate nodes. It should be able to carry on the following continuation of the above dialog.

   > I insist that an animal is a being.
   Then I am inferring that an animal, a being, and a creature
   are all names for the same thing.
   I will use the name animal for this class, and I will
   consider the others to be aliases for it.
   > What is a creature?
   Creature is another name for animal.
   > A creature is an organism.
   I'll remember that an animal is an organism.
   > What is an animal?
   An animal is an organism.
   > An organism is a living-thing.
   I understand.
   > A living-thing is an organism.
   Yes, but a living thing is an organism, because you told me that.
   I'm not going to remember that a living-thing is an organism unless you insist.
   > I insist that a living thing is an organism.
   Then I am inferring that a living-thing, and an organism
   are both names for the same thing.
   I will use the name living-thing for this class, and I will
   consider the other to be an alias for it.
   > Is a being an organism?
   Yes, an animal is a living-thing.
   

3. Optional, for extra credit. (2 points) Extend your Linneus program's functionality to be able to handle the following kind of continuation of the previous conversation.

   > Why is a being an organism?
   Because being is another name for animal, an animal is a living-thing, and 
   organism is another name for living-thing.
   

4. Optional, for extra credit. (5 points) Add memory for what the user entered, as distinct from changes to the main ISA and INCLUDES databases, and the possibility of retraction. Thus you should keep a separate list of the user's inputs, together with corresponding flags, each of which indicates whether the statement has been retracted.

   > A cat is an animal.
   I understand.
   > Forget I said a lion is an animal.
   I have no record that you ever said a lion is an animal.
   > Forget I said a cat is an animal.
   OK
   > Is a cat an animal?
   No, as far as I have been informed, it is not.
   

   (You don't have to handle aliases here.)
5. Extra credit. Retraction of inferrable consequences. (5 points). Add the capability to engage in the following kind of dialog.

   > An A is a B.
   I understand.
   > A B is a C.
   I understand.
   > Is an A a C?
   Yes, it is.
   > It can't be that an A is a C.
   Shall I forget that an A is a B?
   > No
   Shall I forget that a B is a C?
   > Yes.
   OK, it is no longer the case that a B is a C.
   

   (You don't have to handle aliases here.)
6. Optional, for extra credit. Persistence. (3 points). Have Linneus read in a file of facts, isa-data.txt, at the beginning of its session and write out the (updated) file at the end. Ideally, it will also automatically make a backup copy of isa-data.txt with the name isa-data.txt.bak or similar name involving a version number, before it writes out the new version.