武汉理工大学数字图像处理英文课件 Chapter6（9）.pptVIP

Using Planes to Locate Edges Notation Orthogonality Method of Least Squares Least-Squares Solution Locating the Edge Example Case 1 – Centered on Edge Case 2 – Neighborhood 1 pixel to Left of the Edge Case 3 – Neighborhood 1 pixel to Right of the Edge Response to Sharp Edge Weighted Least-Squares Fit Choice of Weights for Linear Edge Sharp Linear Edge Extension to Two Dimensions Skew Edge Example * The equation of a plane is We seek to find the best-fitting plane to a local 3 x 3 area of an image. The elements of the image Z in that neighborhood are shown below The array can be converted to a column vector by numbering the elements as shown. In Matlab, this would be equivalent to the operation Z(:). Next we define arrays containing the local values of x and y over the neighborhood. We assume that the center pixel is the origin and that x increases to the right and y upward. A1=ones(3,1)*[1 1 1]; A1=ones(3,1)*[-1 0 1]; A2=flipud(A1); These arrays can also be converted to column vectors, as shown below. A(:,1)=A0(:); A(:,2)=A1(:); A(:,3)=A2(:); A is a 9 x 3 matrix. We have encountered these masks before and have noted their orthogonality. In this context, the expression of orthogonality is as follows. Note that the multiplications are really dot products. dot(A(:,1),A(:,2)) ans = 0 A(:,1)*A(:,3) ans = 0 C = A*A C = 9 0 0 0 6 0 0 0 6 We can generate a covariance matrix C that summarizes all combinations of the dot products. In the method of least squares, the unknown coefficients are chosen to minimize the mean square deviation of the image values from the fitted plane. The values of the coefficients are found by setting the derivative with respect to each coefficient to zero and then solving the resulting set of equations for the coefficients. The derivatives are given by the following equation which can be simplified by using dot-product notation to where (3x9)(9x1) – (3x3)(3x1) Since the columns of the