What I learned in the first 2 weeks

What I learned in the first 2 weeks Wesna La. Lanne

Edge Detector 1. Find the gradient a. Get a image in gray scale. b. Get the derivative of the kernel in the x and y direction c. Convolve the derivatives of the kernel in the x and y direction with the picture d. Take both convulsions, square them, and add them. Then take the square root of all that e. Boom. Gradient. 2. Pick an appropriate threshold that will show the right amount of details (not too much) so you can get an accurate representation of the edges.

My own edge Detector!

Seagull Example Finding the gradient Detector Gradient + Threshold = Edge

Other things I did with Mr. Seagull Gradient Direction Laplacian

Other things I did with Mr. Seagull Original Picture - Gradient Picture =

Pyramids

Pyramid Edges

Harris Corner Detection ● We use corners because they’re easily identifiable when you look at an image through a small window ● When using corners, shifting said window in any direction, you would see a large change in intensity.

Harris Corner Detection - What’s it doing? 1. Goes through every pixel in the picture to Calculate ‘R’ which is the measure of corner response. a. R = det. M - k(trace. M)^2, where M is a 2 x 2 matrix computed from image derivatives and k is an empirical constant between 0. 04 -0. 06 2. We find the points with large corner response, where R > threshold 3. Take only the points of local maxima R

Box Corner Detection Original Where R > threshold Corner Response R

Lucas-Kanade (Optical Flow) ● Optical flow is a method that is used for estimating the motion of objects across a series of consecutive frames. ● Optical flow has two components: normal flow and parallel flow. Normal flow can be computed directly, but Parallel Flow can’t. Lucas-Kanade is one of several method used to solve the parallel flow issue.

SIFT/SVM Sift - is an algorithm in computer vision that detects local features in the images SVM (Support Vector Machine) - a learning algorithm that analyzes the data from the sift algorithm and recognized patterns.

SIFT in action

Bag of Words ● The algorithm will treat certain features as words and will group the “words” together.