This is it. The fight is almost over. Tomorrow I'm handing in everything and I will be done with my capstone. It feels so unreal, the semester went by so quick and now I'll be graduating on Sunday. So, as you can tell, this will be my last blog post. All I have left is making sure everything is ready to hand in, my website is updated, and I have a binder and flash drive with everything on it. Which won't be too bad, just busy work.

Although this battle has ended, another one has just began. The real world.

Girl ?
​The End ?

Phew. Finally presentations are over and I'm so proud of everything I was able to accomplish and how my presentation went. And to think I was so nervous for it! It went off without a hitch and mostly everything worked as I wanted. Although my k-means merging was random and my labels were one too many sometimes, I was really happy with the final project.

I'm excited to take a couple days to relax and breathe for once, but then I will get back into it for defenses next week and getting my code cleaned up and website done to hand in finally at the end. We have to get the website up and binders with code and documentation and stuff, which will take a little bit but the hard part is definitely over.

Girl 1,000,000
Presentation 0

I'm still currently trying to get my labeling right but over the weekend I was working on coloring the cells when you clicked on them. With this, I had a problem making sure I was only coloring in the cell I clicked on and not all the cells in the image with that same cell number and color. It did take a day to get it but I got it fixed.

This weekend I couldn't focus on the problems I was having with labels because it was crunch time with presentations coming up this week. Mine is on Thursday and I just don't want to miss any of my requirements. So today and the next couple days, I'll be focusing on getting everything to work properly without the project breaking, and most of the requirements done. Right now I'm working on saving the images intermediately so during the presentation, I can show everyone visually the steps that were taken, and then allowing the user to choose the complexity they want and altering the number of cells and colors wanted depending on the answer. 

It will be a hectic couple of days but I'm excited!

Girl 1
Coloring Regions 0

As I said in my last post, this past weekend I started working on getting a specific color palette for my image and making sure each cell was one of those colors. I talked about how it worked in the online project, and I did get it working in my application, but it is just not perfect. Because it using k-means and choosing random colors to see how good the colors chosen were, the final image is different every time. But for some reason when I run mine, the colors are especially off, and a lot of the obvious colors are different. So I will look into that later if I have time.

But, I don't have a lot of time left so I need to work on labeling my images, which turned out to be pretty hard. I implemented a version I found from the online project, and it mostly works. But, the problem I've had is that I'm having a difficult time labeling cells with the same color that are either neighbours to it, or somewhere else in the image. It's very confusing to keep the different groups separate and make sure I'm only putting one label in each cell. So that's what I'll be trying to figure out continuously.

Girl 0
Labels 1

I've been very had at work with all my coloring and merging algorithms, making sure they are working right and not crashing randomly. It took me a couple days to get my cell merging right, but it works great right now and really makes a cool difference in the new image.

In the past couple days, I started to work with the project I found, understanding how it determined the final colors to be used in the image. It seems to be that it uses k-means to grab a random group of colors in your image and determines how good of a group that is. So it does this 30 times and determines the best colors chosen by comparing the score of each group. In the examples this project had, sometimes the colors were different each time it ran, so you got a slightly different image which I thought was cool. So that's what I'll be working on this weekend. And after that is done, I will start to determine how to put labels over the cells.

Girl 2
Color Quantization 0

 So these past two weeks I've been researching different color quantization algorithms that are basically trying to determine the main colors in an image and recreate the original image with only a certain number of colors. The earlier project I found online gave me a really good step-by-step approach that does a couple merging passes to get the best color palette in the image. It groups into cells, pixels in the image that are similar in color, and through the different merging passes, the cells get bigger and it calculates the mean color of that cell.

So far, I've done this first pass grouping pixels into cells, and have been implementing a way to merge small cells and cells with similar colors, but it isn't completely finished yet. Once this is working and done, I need to write a function to get only a certain amount of colors to use in the final product.

Girl 1
Color Quantization 0

This week was very busy because most of us had walkthroughs and as usual, I'm always nervous to present, and I was definitely feeling pretty stuck because my program was not doing what I wanted. I had gotten Sobel's edge detection to work, but I hated how it looked, especially after finding someone's example of a Paint by Number online. The image looked so cool and I didn't think that I could use my edge detection to get to that image. 

So basically, I was feeling pretty overwhelmed and just nervous, not wanting to present something I didn't like to everyone. I ended up visiting with DCP in the morning and then McVey in the afternoon and talk about what I didn't like and what I wanted and they were very supportive. All of us had a thought pop into our heads that said to use edge detection on the image that the example online came up with and after using that image in my program, it worked like a dream! After being pretty excited in McVey's office, I didn't hate what I had. It made me feel a lot better that I didn't think anymore that so much time was wasted on this algorithm.

The walkthrough yesterday went well and I was very happy with what I was able to present. And I have some good feedback and thoughts of where to go from here.

Girl 100
Walkthrough 1

Since I got my grayscale image in my app, I've been talking with DCP and researching Sobel's edge detection, attempting to understand the steps and math involved. But first, to even be able to use the algorithm I would come up with, I needed to get a handle on how WPF (Windows Presentation Forms) dealt with bitmaps and how I could go through the pixels and manipulate them. This proved to be difficult because you need to find a lot of bitmap properties that are pretty vague and it's hard to see how they were connected. 

After a couple of days, I was able to get an implementation of edge detection working but it took a very long time on bigger images, so I knew I needed to find a better way. DCP found an article that gave a different implementation, so I started working on that. I did have to make a lot of tweaks, converting it from the Ruby code it was in to C# WPF, but in the end I got it to work. And this implementation worked a lot faster and had much cleaner results. 

Girl 1
​Sobel 0

For some reason, NetBeans wouldn't admit defeat. Every step of the way, NetBeans fought back, getting more increasingly confusing. Each thing I had to do to get my user interface how I wanted and all the behind the scenes coding was very difficult because I would keep trying things and NetBeans wouldn't act like it was supposed. It took me hours upon hours just to get an image loaded and added as a bitmap to manipulate, and even worse, it was hard enough to even change the font size because when I would increase it, the font would actually get smaller! So I ended up becoming really frustrated and searching online for answers.

​After all those hours and work I put into it, I hated to give up on NetBeans but I knew in the long run, I would still save many hours, especially because I'm getting into the harder stuff, like manipulation my images. So, I ended up switching to C# with Visual Studios because I was more familiar with it and found an alternate way of writing it that it actually had a pretty nice GUI. This ended up being perfect because in one day I was able to rewrite my code easier and get my image to be loaded on the screen.

​The next thing I needed to do was grayscale the loaded image. This was so that when using the Sobel edge detection algorithms, it took less time and calculations as I won't have to deal with RGB values. I was able to do this super easy as I could manipulate the bitmap using some functions.

​So I believe I technically I won against NetBeans.
​Girl 1
​NetBeans.2 0

The blog posts have been forgotten but the fight between Girl and Project rages ever onward. I have made significant progress in my battle as I have decided to do a desktop application using McVey's suggestion of Netbeans, a program with a weird name. It allows me to write in Java, which I am familiar with, and I like a whole heck of a lot better than C# ( gooey-wise. Moving on, I have been working to get used to it, but sometimes it has given me headaches learning how to operate some things and leading me in wrong, harder directions. But alas, I have prevailed and I am still alive and kicking. Netbeans can't get me down.

Today, I met with my sidekick (DCP) to tackle my indecisiveness on how and where to start on altering my picture. We talked about working on edge detection rather than comparing individual pixel RGB colors, which he suggested I start with first, to start on the smaller parts ( He's smart ). Going with that idea, he showed me past students that worked with edge detection algorithms that were similar to what I needed to do ( They smart ). So, my sidekick and these other unaware student sidekicks have helped significantly in my battle.

I must now move forward, ever battling in my fight against my project. All I gotta say, you ask?

Good luck project.
​
Girl  1
​Netbeans Diddlysquat