Week Six
Back to Journal

Week Six: Starting to finish?

This was the week I wanted to "finish up" my code (or rather, shift the majority of my time from writing code to testing it) and I feel like I've done about what I wanted. I realized this week that I could clean up my rotation algorithm even further, and I wrote a tiny piece of code to use the hit-miss transform (more morphology :P ) to remove some artifacts of using Matlab to do the rotation. To Matlab, image rotation is just a matrix transformation, but this simple process leaves sawteeth and other such debris on the edges of objects, so I figured it would be nice to smooth out the edges a bit. It's a minor adjustment, but in some cases it really helps tesseract pick out letters correctly.

I also spent a lot of time this week trying to get my chosen machine translation system, Moses, trained. With statistical machine translation, the computer has to learn translation rules (really, probabilities that words or groups of words in two languages correspond) by observing parallel corpora in the two target languages. I want my prototype system to be able to translate between English and French, and thanks to the U.N. there exist huge, well-translated masses of diplomatic proceedings that my Moses installation will be learning. Since these corpora are so huge (over 10 million words, or almost 1 million sentences!) and the training script is written in Perl, it takes over half a day to train. I was disappointed to see after over half a day running on my home machine (murdoc, a Toshiba laptop with 2G of RAM - not too shabby) the process crashed due to memory constraints, so next week I will be inflicting some language-learning pain on the Princeton servers instead. Let's hope they can handle it better than poor murdoc!

Margaret and I had our flexibly-scheduled weekly meeting this Thursday after she returned from an architecture conference in Beijing. (On a non-technical note, the Olympic buildings sound awesome, but the soon-to-be Olympic smog sounds less so. Apparently the view gets hazy even a few blocks down a street!) We've decided I'm going to try and get two papers out of this project - one conference-length paper on my entire image preprocessing system and its context, and one shorter focus on my text rotation algorithm. I am less confident about the latter - since I've gone through the paper-reading and several implementations step by step, my algorithm seems a bit derivative to me - but she thinks my approach is worth documenting since no one's taken it quite this way before. Now with two papers on my plate, I have to get cracking on testing this thing! I see a lot of OCRed phonecam images coming next week...

My friend Catie from GT came to visit this weekend - it's really great to have visitors! I showed her around Princeton on Saturday, which didn't take long since it's pretty small, and around Hopewell and my house. We're also quite into movies, so we went to see Get Smart - much funnier than I expected! No baking over the weekend, but I made some blueberry muffins from Terhune's muffin recipe on Wednesday. (That had to take the place of So You Think You Can Dance, because Erika was having a meeting in the living room for the school she and some other teachers are starting.) The tofu cheesecake is setting in the refrigerator, waiting for my sister and I to test it out this weekend. That might be the weirdest thing I've ever baked, but I swear it's a good idea - the best cheesecake I ever had was in Japan and they do theirs with tofu!