Week Seven<title> </head><body> <div class="all"> <div class="content"><span class="title"><br> Week Seven</span> <br> <a href="journal.html">Back to Journal</a><br> <br> <span class="bold">Week Seven: Seeing Results</span><br> <br> Though this week marked the start of my transition to testing, I started the week with a few meetings. On Monday, Margaret invited Sahar, the other DMP student, and I to have lunch with Jennifer Rexford, a professor in the CS department at Princeton. This was a particularly interesting meeting for me - the professor I work with back at Georgia Tech, Nick Feamster, has worked extensively with Dr. Rexford, so I've read a few of her papers in preparation for my own work. I enjoyed our informal little lunch - like so many of the faculty members I've met because of research, Dr. Rexford is fun and interesting on top of her intellectual merits - and I look forward to seeing her at events like SIGCOMM in the future.<br> <br> I also had a meeting with several members of the SARANA project, a space- and resource-aware networking platform in which Margaret and her students are involved, on Tuesday. My current project is not immediately for SARANA, but we started it for its application's and development on Android's potential use in their network. It was very enlightening for me to hear what resources they need for SARANA's extensions to the Java programming language, and we started to think of how the Android platform can meet these needs. One of my lower-priority (behind my research papers!) tasks for this summer is to pull together some documentation of my experience with Android so it doesn't leave Princeton with me, and after our meeting I feel much better in tune with how I can contribute to their project.<br> <br> By Tuesday afternoon, I had prepared enough for meetings that I was ready to get back to my own research. I started with a few last ideas about how to improve the text detection algorithm I'm using. An early concern of mine was shadowing - how, in an image mottled with shadows, can you tell what edges belong to letters (or other foreground objects) and what belong to changes of light on a solid background? The Hasan/Karam algorithm for detecting text is fairly robust since it uses morphology to group edges before thresholding (picking out objects based on whether they surpass a threshold level of light or dark), but many papers I've read suggest taking thresholding a step further by picking the threshold levels in local regions instead of across the entire image. This would mean that areas of shadow would not have so much bearing on what threshold is used in areas of normal light. One local (or adaptive) thresholding method of interest is Niblack's method, which simply uses the mean and standard deviation of grey values in each small section of the image and picks a threshold level for that region based on a weight assigned to the standard deviation (ie. anything k standard deviations below the mean is a text pixel). I did a quick implementation of this algorithm in Matlab, and honestly it doesn't seem to be helping much. In most cases, even what I would consider extremely non-uniform shadows, a global thresholding does the job after the morphological steps.<br> <br> On Wednesday, I gathered my test data and wrote the script to process all my test images. I want to test both black and white documents (I think they are a good sort of control - nothing confounding to the morphological or thresholding operations) and scene text (mostly from street signs). My document images are passages in English and French from The Little Prince - I enjoy having to read them, and I know enough of both translations to see if Moses, the statistical translation system, performs well enough on it (if it ever trains up and I can use it - the training process takes ages!). Getting the scene text turned into a full blown reconnaissance mission - I walked all up and down Nassau Street and through campus a little on the hunt for different signs. (I ended up at The Bent Spoon for two scoops of sorbet before returning - had to keep my energy up! ^_~) Now I have about 36 street signs and some few dozen pictures of document text - and I hope to expand this corpus once I've analyzed the initial results a bit - and my tests show a lot of potential. Nothing's perfect yet with the scene text, but I feel like I've gotten some ideas for improvement that should be implemented this coming week.<br> <br> I had a lot of fun last weekend when my sister came to visit. We've had to deal with being apart for awhile before - she was in England for all of fall semester last year, so we got used to calling a lot - but it is always nice to chill with my best friend in person. We are movie people, so we had to see two movies - WALL-E was adorable (and eco-tastic; what a great medium for bringing kids into the environmentalist fold! har) and I liked The Hulk more than I expected (Edward Norton is so good at playing the tortured hero). The cheesecake, sadly, was not "our thing" (though it was my housemates' thing - it got mostly eaten before I threw it out late this week), so I had to make another dessert this weekend. I did take off for the Fourth of July (I had thought I might go into work, since I have a lot to do for these few papers, but that was a silly thought! It's a national holiday!) and made a blueberry pie that definitely has the title of Best Pie Megan Made in New Jersey (so far! there's a few weeks left). Erika and Sharad were also around on Friday, so after licking our blueberry pie plates for awhile we drove my car up a hill in Hopewell and watched the surrounding fireworks displays in high style (sitting in the car because of the rain and hollering each time we saw one so everyone would turn in the right direction). I had wished to spend the holiday with my family or friends, but it ended up to be a really satisfying, simple day that reminded me of how glad I am to have ended up renting with such nice people. </div> </body></html>