Exercise 9 - Background Subtraction

This exercise involved taking out the background and leaving the foreground of the video. Our lab instructor had already provided the code, but the output of the said code was a black foreground. Our exercise involved producing the actual foreground, and on a video output file.

While we weren't able to finish the exer, we did understand how background subtraction on videos work. The video will have to be spliced to frames, then each frame would be studied for a possible background, then the studied background is subtracted to all frames, then the processed frames are made to a video again.

Exercise 8 - Color Tracking

This exercise was about false coloring, and we were required to apply our own color scale on the sea part of the provided images. It was fairly easy, as we used the inRange() function for identifying the blue waters part of the images.

Exercise 7 - OCR Part 1

For this exercise, we were made to produce 20 images of handwritten numbers, with one number per image, then and another 20 images of computer-made numbers, then using our own implementation of OCR, we must be able to study and learn the features of the computer-made numbers, then identify the numbers written on the handwritten number images. The determinant for identifying the numbers was using a Euclidean distance formula. We used Area, Height, and Width for the exercise.

While our exercise worked without much error, we were only able to identify 25% of the images. It was still good, despite that we already had used 3 features for the exercise. Some identifications were really far (like 3 identified as 7) and some are close (like 9 identified as 0).

Exercise 6 - Erosion-Dilation

This exercise is about using image erosion and image dilation techniques in order to fix up some sample student evaluation forms, then tally the results. What we did was binarize the image first through thresholding, then experiment on using the erode and dilate to emphasize the marks that were shaded. We mapped the individual coordinates first, then tally the results and record them on a text file.

Exercise 5 - Blobs

In this exercise, we had to (1) identify the coins and compute for the total amount of the coins on the photo provided, and then we had to (2) identify the images on another photo.
While we did not follow the instructions for labeling the coins and images (LOL), we were able to identify the coins present on the first photo, as well as the objects on the second photo. We had a bit of problem when it came to OpenCV versions, as some functions work differently on some versions, but it wasn't that big of a problem in regards to doing the exercise.

Original coins image.

 Labeled coins, they should be labeled using bounding boxes.

 Original objects image

Labeled image

Exercise 3

The exercise that week was about Filtering images. The exercise required us to make a medianBlur filtering function. The exercise was rather easy, however my partner and I had some troubles.

First off, the function we made only catered to kernels with only 1 as their values. The function wouldn't work well on kernels with different cell values. Second, because of the first problem, we couldn't test other kernels such as the Gaussian Kernel.

While we both learned how filtering works, it was such a shame that we weren't able to do the exercise correctly.

Exercise 4 was rather a mix of easy and hard, unlike the other exercises. The exercise was all about binarization and edge detection. Binarization is a process of turning the photo foreground to black, then the background would be white, or vice versa, given a threshold. Edge Detection simply means detecting the edges or outlines of the foreground of the photo.

Now why did I thought it was a mix of easy and hard? That's because the difficulty of these two processes depends on the photo one is working with. For example, in the Binarization part, we found the quote.jpg and magazin1.jpg easier to binarize, while magazin2.jpg was harder to binarize. My partner and I used brute force to get rid of the annoying background so it would be easier to make the text appear.

Original quote.jpg.

Binarized quote.jpg where the the letters of the image are all in black.

Original magazin1.jpg.

Binarized magazin1.jpg.

Original magazin2.jpg.

Binarized magazin2.jpg. Notice how the photo is still not clean and still has the noise from the original photo.

As for the Edge Detection part, the photo was really hard to work on with, since one part looked illuminated while the other didn't. While I was able to (somehow) get the edges, the algorithm we did may not be correct as we still used brute force to binarize the photo.

Original source.png.

Binarized source.png.

While I understood the purpose of Binarization, I didn't appreciate the Edge Detection much, since it invloved a lot of steps compared to Binarization. I still understood the two concepts nonetheless. Hopefully I would get to appreciate Edge Detection when we get to the later exercises.