Week 4

Week 4

My goal for this week is to finish the “mini” functions (co-occurrence matrix, normalized co-occurrence and the five texture measure for small windows of interest within larger images.) Once they work well, then I need to start thinking about a classification system for analyzing the effectiveness, usefulness of my texture measures.

Along with the mini functions I am developing, I also need to implement two functions, one for putting a border and one for taking away the border from an image. Convolution involves the multiplication of an large matrix (image) with a small matrix (mask). The convolution is multiplying the small matrix with similar sized pieces of the image. The end product depends on the mask, common convolution masks in image analysis yield the gradient of the original matrix. Often the border of the image is lost in the convolution process. By adding a border that is half as wide as the “pieces” of the image that are being analyzed at a time, the mask can be applied to the entire image. The resultant image would have a border that has no useful information, so we must delete the border before moving on.

Also once I have my mini functions ready, I need to make a function that calls each of the functions in turn and returns the numeric value of the texture measures. The process must be done for the entire image but taking 40 by 40 pixel windows with a 20 pixel horizontal and vertical overlap.

This week is the 4^th of July and I have been invited to a bbq at the house of one of the grad students. Apparently It has a very good view of the fireworks at Gas Works park.

Also this week is Anjali’s b-day, she’s one of the DMP students working in the EE building. She wants to celebrate her b-day at a restaurant downtown. Also in attendance will be other Berkeley students on internships at Microsoft, and Amazon. They are classmates of Anjali.

This week went by very fast. I accomplished my goal. The algorithm that computes the texture measure in small windows was the most challenging, not so much in the implementation, but in the design, putting all the pieces together. I made two versions of that program that I will present at next week’s Tuesday meeting. Either I can compute a 5 dimensional matrix that in each of its positions stores the five values corresponding to the texture in that window, or I can compute 5 matrices, each storing the values for a specific texture measure. The most challenging part of the implementation was representing the size of the output matrices in advance eof running the program. I kept being off by one, until I used the ceiling function.