340 likes | 735 Views
Computer Vision – Enhancement(Part III). Hanyang University Jong-Il Park. The Fourier transform. Definition 1-D Fourier transform 2-D Fourier transform. Fourier series. 1- D case. M-point spectrum. 2 D Fourier series. 2-D case is periodic : period = 1
E N D
Computer Vision – Enhancement(Part III) Hanyang University Jong-Il Park
The Fourier transform • Definition • 1-D Fourier transform • 2-D Fourier transform
Fourier series • 1-D case
2D Fourier series • 2-D case • is periodic : period = 1 • Sufficient condition for existence of
Eg. 2D Fourier transform original 256x256 lena Centered and normalized spectrum (log-scale)
Unitary Transforms • Unitary Transformation for 1-Dim. Sequence • Series representation of • Basis vectors : • Energy conservation :
2D Unitary Transformation • Unitary Transformation for 2-D Sequence • Definition : • Basis images : • Separable Unitary Transforms:
Unitary transform Point operation Inverse transform LPF LPF BPF BPF LPF HPF BPF HPF BPF BPF LPF LPF Generalized Linear Filtering • Generalized Linear Filtering Zonal masks for Orthogonal(DCT, DHT etc) transforms Zonal masks for DFT
Ideal LPF NOT practical because of “ringing”
convolution Illustration of Ringing Ideal LPF
Eg. 2nd order Butterworth LPF A good compromise between Effective LPF and Acceptable ringing
Eg. GLPF No ringing!
Application of GLPF(2) Soft and pleasing
Homomorphic Filtering • Homomorphic Filtering • f(x, y) = i(x, y) • r(x, y) i(x,y) : - illumination component - responsible for the dynamic range - low freq. Components r(x,y) : - reflectance component - responsible for local contrast - high frequency component enhancement based on the image model - reduce the illumination components - enhance the reflectance components
Linear System log exp LP exp log <1 f(x, y) HP g(x, y) >1 Transform Operations • Homomorphic System note