Weekly Update for April 9

Sample Slide

A sample list:

1. O-Ren Ishii
2. Vernita Green
3. Budd
4. Elle Driver
5. Bill

Nist Scans: Type II Error rate and Failed Tests

Book Progress & Modeling & Writeups, Oh My!

1. Chapman and Hall book: Progress! Did a lot of research and some writing. Rounding out the chapter this week.
   
2. Grooves Writeup: Started a write-up of methods we are trying for groove identification
   
3. Modeling: Trying out a new method. First predicting whether a groove is there or not!
   
4. Hands-On Stuff: Going through and identifying where there are grooves/where there are not for test set.
   
5. Reading Group Talk!: On NIJ Fellowships. Get excited!
   
6. Plots coming next week… :)

Bivariate SLR Weirdness

• score(x,y) = (|x - y|, |y|)
• x is the known source measurement
• y is the unknown trace measurement

Bivariate SLR Weirdness cont.

Writing away

• Glass chapter
• Thompson redo
• Judge training
• REU plans, lectures, assignments, etc.

Shoeprint registration in MATLAB

• Made it work after changing affine transformation to rigid transformation (translation and rotation only)
• The alignment is good enough to make comparisons. Not perfect just due to differences in the images
• Working right now on writing a script to align all images to date images per size and model

Example

Cleaning

Stentiford, Fred and R. G. Mortimer. “Some new heuristics for thinning binary handprinted characters for OCR” IEEE Transactions on Systems, Man, and Cybernetics SMC-13 (1983): 81-84.

• Implement pre-processing from the paper above.
• 13 steps based on masks
• Goals: Remove “spurrious projections”, fill “noise holes”, limit “necking effect”, eliminate “acute angle tails”
  • Current processing fills “noise holes” only
• All of this is done with the goal of making processing more robust so differences in thinned writing come from writing, not processing.

Handwriting

• Latent Dirichlet Allocation (LDA) - Thanks Danica! :)
  
• Results from modeling next week!
  
  
• Next: Nick’s graph counts & next level data!