I. Introduction

II. The Steps
	a. Conversion (show image)
	b. binarization (show image)
	c. thin (show image)
	d. minutiae extraction
		1.what is a minutae point
	e. process - translation and scale invariant
	f. Nearest Neighbor

III. Testing
	a. first used fingerprint generators

	b. the database
		1. From the Dept of CS at the University of Bologna 
		2. 160 real fingerprint images
		3. 8 prints of one finger from 20 different people
		4. all captured using a scanner 
		5. 256x256 pixel Tiff files
		6. all seem to be vertically oriented

	c. compared the first print of each set of 8 to see what sums 
	were returned when matching prints where compared

	d. Compared the first print of each set to the first print in each 
	of the other sets to see what sums were returned when the prints 
	compared did not match.

	e. using this information chose some cut of values for a match
		1. 3.5
		2. 3.0
		3. the average of the sums returned when the first print 
		if each set was compared to the other members of that set
 
	f. re-evaluated the sums recived from the earlier comparisons in 
	light of the cut off values to see what pairings of prints would 
	be considered to match and which ones were not

	g. relaized that I was not happy with the results and I wanted to 
	tinker with it some so I tried the test again this time without 
	pruning the images

	f. from this testing I was able to determine the false rejection 
	and false acceptance rates
		1. the false rejecion rate is the number of times a print 
		that matches is said not to match. thus this is the rate 
		at which a person who should have access is denied

		2. the false acceptance rate is the rate at which prints 
		that do not match are said to match  

		3. the rates found by my program

			A. Read what the rates are for the pruned and 
			non-pruned images

			B. shows that some matching is being done
			
			C. Have to mention that the nearly 50% false  
			rejection rate using the averages is not unexpected


		4. What is optimal depends on the level of security 
		wanted and the level of user convience expected

			1. what I mean by convinience is how often are the 
			users willing to be turned away 
			
			2. while using the averages in both the pruned and 
			non pruned and using 3.0 would let in the fewest 
			impersonators, they would also reject authorized 
			users often.  

			3. Meanwhile, using 3.5 without  pruning would 
			deny authorized users the least, it would also 
			let in a number of unauthorized users 

			4. personally like averages without 
			pruning the best. (3.0 no pruning is good but 
			sometime completely denied access to some people 
			because the sums returned when matching prints 
			were compared was high for those prints ***this 
			could be fixed during enrollment)

		
IV. Comparisons with other programs

	a. pretty pleased that it did anything

	b. other programs reserch do a much better job of fingerprint 
	verification.

	c. some of the top results from three other programs that have 
	been studied

	d. read their results

	e. note that filterbank is not minutiae matching algorithm



V. Improvements 
	
	a. the results shown by the other projects give something to 
	aspire to.

	b. I found several places that could use improvement

	c. figure out why so many minutiae are being found
		1. in prunned 123-794
		2. not pruned 579-1779
		3. supposed to be 60 - 80
		4. my hypothesis is that picking up pores, wrinkes in 
		skin, scars
		5. may be fixed with a smoothing function, 
		6. better  binarization may help
	
	c. why it works better without pruning is another question I would 
	look into.  
		1. My tenative thought is that the real fingerprint 
		images have more broken ridges that are short and easily 
		erased taking away some of the data needed do the matching
		
		2. smoothing may help here as well

		3. better pruning methods or tinkering with the ones used, 
		as I have not exhausted the possible arrangement of the 
		pruning elements

	d. making the print rotationally invariant.  

		1. this was not done because it is complex and time 
		consuming

		2. couldn't find a good algorthim

		3. fortunatly the prints used in this study seemed to be 
		vertically oriented, but having functions to do this would 
		make the system more robust
	
VI. Conclusions
	
	a. learned a lot (didn't know anything before)

	b. exposed to image processing which was another topic I was 
	interested in