Features and object recognition

Question 1

What is a feature?

Answer 1

• We’ve described features as points that are easily compared and tracked among images of the same object or scene.
• Corners is the most important feature because it is the most repeatable feature and are more unique

Question 2

Types of features?

Answer 2

• edges: areas with high-intensity gradient or high changes in intensity
• corners: an intersection of 2 edges
• blobs: region based features. areas of extreme brightness or unique texture

Question 3

important functions

Answer 3

• cornerHarris corner detection
• dilate to enhance areas
• Erosion to decrease areas
• morphologyEx to open or close
• fastNlMeansDenoisingColored reduce noise

Question 4

corner detectors

Answer 4

• A corner is an intersection between edges. So a corner can be found when there is a high gradient in many directions. Gradient has magnitude and direction. To calculate it use sobel x and y operators
  In a gradient image
  1) Shift a window around an area in an image
  2) Check for big variations in gradient magnitude and direction

Question 5

Dilation and Erosion

Answer 5

• Dilation enlarges bright, white areas in an image by adding pixels to the perceived boundaries of objects in that image. Erosion does the opposite:

Question 6

Opening

Answer 6

• erosion followed by dilation. This is useful in noise reduction in which erosion first gets rid of noise

Question 7

Closing

Answer 7

• Closing is the reverse combination of opening; it’s dilation followed by erosion, which is useful in closing small holes or dark areas within an object.

Question 8

Features vectors

Answer 8

• Fundamental for object recognition
• Divide image into squares and calculate the gradient magnitude and direction from the center of the image for each segment

Question 9

HOG

Answer 9

• Algorithm for image recognition using gradients
• uses histograms to organize image data!
  1) Calculate magnitude and direction of gradient each pixel
  2) Group these pixels into square cells
  3) Count how many gradients in each cell fall into each orientation

Question 10

Features classifier models

Answer 10

• Convolutional neural nets
• Support vector machines
• Decision tree

Question 11

haar cascades

Answer 11

• Is an algorithm for classification
• Haar features to detect edges, lines and rectangular patterns
• Cascade of Classifiers discards parts of an image that are not the object of interest

Question 12

optical flow

Answer 12

• Used in video and motion analysis.
• Used to predict the future location of a point. Applications for hand gesture recognition or tracking a moving vehicle
• Assumes that pixel intensities stay consistent between frames and that neighboring pixels have similar motion
• Optical flow uses object features and an idea of perceived movement to track points and objects in consecutive frames.