engre 133 checkpoint 2

1. How are Sobel kernels different from Gaussian kernels?

Gaussian - all positive weights and blurs the image

Sobel kernels - have negative and positive values and used to measure change and highlight edges

2. What is the x-gradient image?

It highlights vertical edges. Each pixel shows how much its change from left to right

3. What is the y-gradient image?

It highlights horizontal edges. Each pixel shows show much its values change from top to bottom.

4. What is meant by gradient?

The rate of change of pixel values in a given direction.

5. What is the Sobel filter detecting?

Detects edges, places where pixel intensity changes sharply

6. Why do we apply a threshold?

A threshold makes each pixel strictly an edge (255(white)) or not an edge (0(black))

7. What happens if the threshold is too high? What happens if the threshold is too low?

Too high - too few pixels count as edges so real edges and detail are lost

too low - too many pixels count as edges so the edges are cluttered and detail is lost

8. Why do we want to detect the edges of the image?

Edges can capture the outlines and shapes in the sign and strip away color and brightness. We use those shapes to tell signs apart and extract its features.

9. What features are extracted from the color image?

the mean and standard deviation of each H,S,V values

10. What features are extracted from the grayscale image?

count of lines detected, if it have circle, if it have triangle

11. What filter(s) do you use to detect circles?

gaussian and hough circles

12. What filter(s) do you use to detect triangles?

gaussian filters, sobel filters to get edges, and contours to find the shape, if it have three corner, triangle

13. What filter(s) do you use to count lines?

sobel + hough lines

14. What is the metadata CSV file? What data does it contain?

The input file listing every image in the datasets, contains path and classID, and it tells the program which images to process what each one is

15. What is the purpose of extracting features from our images?

They turn images into a numbers that machine learning models use to tell the classes apart…use this to train and test the classifier in next checkpoint

16. What is the features CSV file? What data does it contain?

The output with one row per image and 11 columns which is the 9 features plus path and ClassID

17. What is feature scaling? Why is it useful?

It puts features on the same scale so none dominates. Useful for features with big ranges it don’t dominate features with smaller ranges.

18. What is the feature space? Why is it useful?

A space where each axis is one feature and each image is a point at its feature values. It shows whether the classes form clusters and which features separate them best.

19. How do the feature plots change when you apply feature scaling?

they will have the same shape and different axis numbers center at 0

20. Why do the plots with has triangles and has circles look so different to the plots comparing the color features?

They give outputs only 0 or 1 stack in a few spots. but color features are continuous spread out and form clusters.