Skip to content

Latest commit

 

History

History
88 lines (73 loc) · 3 KB

Task3.md

File metadata and controls

88 lines (73 loc) · 3 KB

Task 3

Haris Operator

std::vector<_Point> cornerHarris(Image& grayImg, double threshold, int patchDim = 3)

Parameters:

  • Grayscale image
  • Threshold(0 -> 1) which is percentage of max of haris operator
  • Patch dimension to construct hessian matrix at each pixel
    it uses internally void getHarrisCorner(Image &pointsStrength, std::vector<_Point> &cornerPoints, double threshold) which get the points that greater than threshold
    Complexity: O(imgWidth x imgHeight)

SIFT Descriptor

std::vector<std::pair<std::vector<double>, _Point>> getSIFTDescriptor(Image &inputImg, float threshold = 0.01)

Parameters:

  • Grayscale image
  • Threshold(0 -> 1) which passed to cornerHarris

Returns: vector of pair, each pair if 128-length vector and its associated point

Steps:

  • get main orientation of pixel std::vector<double> getMainOrientation(Image &dir, Image &magnitude, std::pair<int, int> iRange, std::pair<int, int> jRange) construct histogram of 36 bin in neighborhood of the pixel and take all orientations that greater than 0.8 * (max orientation)
  • for each main orientation construct 16 x 16 neighborhood for each 4 x 4 region make histogram of 8 bin (of relative orientations) and store this histogram in the feature vector void featureHistogram(Image &dir, Image &magnitude, double mainOrientation, std::pair<int, int> iRange, std::pair<int, int> jRange, std::vector<double> &v)
  • normalize the feature vector

Complexity: O(n^2)

SSD Matching

std::vector<double> SSDMatching(Image& inputImage_1, Image& inputImage_2);

Parameters:

  • Grayscale image (first image)
  • Grayscale image (second image)

Returns: vector of points that match

Steps:

  • get FIST descriptor of both images
  • loop over the descriptor of the first image and inside this loop , loop over the descriptor of the second image
  • calculate the SSD between the key point of the first image and all the key points of the second image
  • get the 2 least SSD points and make sure that the least point is less than 0.8 * the second least point
  • put the points in the vector
  • return the vector holding the points

Complexity: O(n^2)

Normalized Correlation Matching

std::vector<double> normalizedCorrelation(Image& inputImage_1, Image&inputImage_2);

Parameters:

  • Grayscale image (first image)
  • Grayscale image (second image)

Returns: vector of points that match

Steps:

  • get FIST descriptor of the 2 images
  • loop over the FIST descriptor of the 2 images and get the average of the vector of each key point 128 vector and store it
  • loop over the descriptor of the first image and inside this loop , loop over the descriptor of the second image
  • calculate the Normalized Correlation between the key point of the first image and all the key points of the second image
  • put the points in the vector
  • return the vector holding the points

Complexity: O(n^2)

Result