P. 1
Noise

|Views: 3|Likes:

See more
See less

03/24/2012

pdf

text

original

# PAM3012 Digital Image Processing for Radiographers

Image Restoration
• Image Enhancement • Image Restoration
– Improve appearance of image – Subjective – Recover image after degradation – Objective

Noise & Image Restoration

• Concentrate on Removal of
Noise

Image Restoration
• Some restoration techniques are better
formulated in spatial domain and others in frequency domain

• To restore an image we must model degradation
process so that reverse process can be applied

• Spatial

• Frequency

– Periodic Noise – Blurring

^ • Objective of restoration is to obtain estimate of f(x,y) f(x,y) • Given g(x,y) and some knowledge about H & η(x,y) ) ) g(x,y x,y

Noise Models
• Principle sources of noise in digital images arise
during acquisition and/or transmission
– Sensor noise – Transmission – Display

• Representation in frequency domain

• Noise can be characterised by it’’s Spatial & it
Frequency Properties

G (u ,v ) = H (u ,v ) × F (u ,v ) + N (u ,v )

1

y) – Histogram of noise component Gaussian Noise • PDF of Gaussian noise • Common PDF found in image processing… processing… – Gaussian and Impulse 2 p (z ) = e 2π σ − ( z − μ )2 2σ 2 • z = grey-level grey• μ = mean value of z • σ = standard deviation Impulse Noise • PDF of Impulse noise ⎧Pa ⎪ p (z ) = ⎨Pb Effect of Noise on Image Noise Free Image Gaussian Noise Impulse Noise ⎪0 ⎩ for z = a for z = b otherwise • Also referred to as salt-and-pepper noise salt.y) x.and• Impulse noise is generally large – often digitised as black & white values in image Periodic Noise • Electrical or mechanical interference during acquisition Estimation of Noise Parameters • Periodic Noise – Inspection of Fourier Image • Determining PDF • The only type of spatially dependent noise – Sensor specifications – Inspection of histogram of ‘flat’ image flat’ – Determine shape of Gaussian by measuring μ&σ 2 . η(x.Noise Probability Density Functions • Statistical behaviour of grey-level values in greythe noise component of image.

v ) Now easier to represent g (x . y ) + η(x .v ) + N (u . Gaussian or Impulse Restoration in the Presence of Noise Only – Additive Noise 1 R = Spatial Filtering mn • Mean (or Box) Filter ∑z i =0 nm i • Mean Filter • Median Filter Image corrupted with Gaussian Noise Restoration in the Presence of Noise Only – Additive Noise • Median Filter Spatial Filtering Restoration in the Presence of Noise Only – Periodic Noise • Band reject Filter Frequency Domain Filtering Image corrupted with SaltSaltand-Pepper Noise and- 3 . y ) = f (x .v ) = F (u .v ) × F (u .v ) + N (u .v ) Becomes G (u .e.v ) = H (u .Restoration in the Presence of Noise Only Restoration in the Presence of Noise Only Restoration Processes: • Noise only therefore no degradation term: G (u . y ) in spatial domain • Spatial Domain Filtering – Additive Noise (Gaussian & Impulse Noise) • Frequency Domain Filtering – Periodic Noise Restoration in the Presence of Noise Only – Additive Noise Spatial Filtering • Method of choice when only additive noise is present – I.

Restoration in the Presence of Noise Only – Periodic Noise • Band-reject Filter BandFrequency Domain Filtering Restoration in the Presence of Noise Only – Periodic Noise • Notch Filter Frequency Domain Filtering 4 .

scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->