Week 2

This week I moved onto implementing texture computing algorithms.

Last week was about getting oriented. My mentor played a pivotal role last week. Introducing me to other graduate students and visiting scholar that I will be working with. She told me about a talk by a leading researcher in the field of computing vision. She took me to eat to the faculty club on my first day. I had no idea that it could feel disorienting, but it makes sense, I will be working with so many different people: the support staff (whose office I’ve been to more than once already), the other graduate students in the vision group, my graduate student office mates, the graduate adviser who gave me a tour of the school and introduced me to a few people on my first day, other graduate students in the building, the receptionists down at the computer science office … the school itself is big, the are to me is unfamiliar, but by the end of the week, I had made several acquaintances, including two other DMP students working in a lab in the Electrical Engineering building, which is connected to the computer science building.

They have a very good social environment here at the University of Washington’s computer science department. Fridays at 10, there is a breakfast. Different graduate students take turns bringing the breakfast and many graduate students get together and talk about life, their projects … it’s just a casual time to meet others and build a sense of community. The graduate adviser told me about the breakfast, I invited the other two DMP students. A professor also joined us for breakfast and I felt welcome.

I have confidently eased, and adjusted to my surroundings. I know where to find help, and where things are located, generally. I moved into my place, it’s a small studio two blocks from campus. I walk to work and so far I’ve gone out to lunch with other grad students or DM students every day. We go to the Ave, which is actually University Way and has many restaurants.

I’ve started attending the weekly Tuesday morning Vision group meetings. My mentor, Linda, heads the vision group. It is comprised of students, a post doc, and a person who already has her PhD but still comes to the meetings to report on her after PhD adventures. The graduate students are at various levels of their PhD program. Habib is getting ready to defend his dissertation, also Raven and Jill are getting ready but will defend about a month after Habib. Enrique is getting ready to do his Thesis, Marta is a visiting scholar from Spain doing a six month post-doc. There are other members, some are in internships in industry. The meetings are casual, people take turns talking about their work, their progress, any problems they are encountering, and the group suggests possible solutions. The atmosphere is inviting and has a positive tone, so that other peoples suggestions don’t feel like negative, but rather as alternatives so as to overcome ones problems.

It was a short week for me, because Thursday night I flew back home to San Diego to attend my sister’s graduation.

In the 4 days I was here I started implementing algorithms to compute the co-occurrence matrix. The co-occurrence matrix stores information about uniform regions, as well as local transitions (edges) in the input image. However, the co-occurrence matrix alone is not directly useful for further analysis. Instead, numeric features are computed from the co-occurrence matrix that can be used to represent an image’s texture more compactly. We first compute the normalize co-occurrence matrix since we are comparing different bit-depths and so our measures must range in the interval [0,1]. Once the normalized co-occurrence matrix is know, we proceed to compute the five standard features: Energy, Entropy, Contrast, Homogeneity, and Correlation.

One of my stumbling  blocks as I implemented the co-occurrence matrix algorithm in Matlab was having the right data type. I used the imread function to store a gray scale image in an array, but before proceeding I had to modify to converting that array to double data type. Also there was the question of, what if my image is not a gray scale image? So I added a line of code that first converted whatever image I received to gray scale, if the image was not already gray scale. Due to y unfamiliarity with Matlab, I didn’t take advantage of many of Matalb’s built in matrix manipulation features. I need to debug my algorithms, namely, the co-occurrence, and the algorithm that finds the correlation. I am not sure what the problem is. I am testing my algorithm against Matlab’s co-occurrence and feature derivation functions. The idea was for me to familiarize myself ‘from the ground up’ with the texture features, as they will be pivotal parts of my work in the coming weeks.