全面的图像滤波方面课件.pptVIP

Odd Gabor filter First Derivative Even Gabor filter Laplacian Separable, low-pass filter Not-separable, approximated by A difference of Gaussians. Output of convolution is Laplacian of image: Zero-crossings correspond to edges Separable, output of convolution is gradient at scale s: Gaussian Derivatives of Gaussian Directional Derivatives Laplacian Fourier Transform F(g)(u,v) = component of image at frequency sqrt(u2+v2) in direction (u,v) Transform of Gaussian s is Gaussian 1/s Output of convolution is magnitude of derivative in direction q. Filter is linear combination of derivatives in x and y Steerability(可操控性) Generalization of property of derivatives: F is steerable if the rotated filter can be expressed as a linear combination of basis filters. Gabor Filters Compute the local contribution of frequency in the direction (kx,ky) at scale s. For s large compared to 1/f, even filters approximate 2nd derivative, odd filters approximate 1st derivative. Special case of Gabor wavelets if s and f restricted to powers of 2 Gaussian Pyramids Gaussian smooth image and subsample at each stage Compute Laplacian by difference of Gaussian at each stage Laplacian Pyramids Scale-Space Convert the image f(x,y) to a continuous function of scale Directional smoothing Smooth with different scales in orthogonal directions Filtering Basic Filters Convolution/Linear filtering Gaussian filters Smoothing and noise reduction First derivatives of Gaussian Second derivative of Gaussian: Laplacian Image pyramids Oriented filters/frequency analysis: Oriented Gaussian Oriented derivatives Fourier Transform Gabor filters Steerability What is noise? s = 1,3,5 Gaussian filter s = 2 s = 4 Gaussian smoothing p1 p2 p1 p1 p1 p2 P2 P1 G Gx Gaussian and its 2 derivatives LOG operator Local image characteristics based on: Scale Frequency Direction q Fourier basis Spatial domain Frequency domain 1/s 1/s s s Even (Symmetric) Odd (antisymmetic) Gabor Filter Example High fre