Image Enhancement To poce an image o that output i viually bette than the input, fo a pecific application. Enhancement i theefoe, vey much dependent on the paticula poblem/image at hand. Enhancement can be done in eithe: Spatial domain: opeate on the oiginal image g(m, = T[f(m,] Fequency domain: opeate on the DFT of the oiginal image G(u,v) = T[F(u,v)], whee F(u,v) = F[f(m,], and G(u,v) = F[g(m,],
Image Enhancement Technique Point Opeation Mak Opeation Tanfom Opeation Coloing Opeation Image Negative Contat Stetching Compeion of dynamic ange Gaylevel licing Image Subtaction Image Aveaging Hitogam opeation Smoothing opeation Median Filteing Shapening opeation Deivative opeation Hitogam opeation Low pa filteing High pa Filteing Band pa filteing Homomophic filteing Hitogam opeation Fale coloing Full colo poceing
Point Opeation Output pixel value g(m, at pixel (m, depend only on the input pixel value at f(m, at (m, (and not on the neighboing pixel value). We nomally wite = T(), whee i the output pixel value and i the input pixel value. Outut gayvalue (nomalized) 0 0 Input gayvalue (nomalized) T i any inceaing function that map [0,] into [0,]. Matlab Aide: Thi i oughly what matlab doe when you convet a uint8 image to double a = imead( image.tif ); b = im2double(a);
a i an uint8 aay, with gavalue in [0, 255]. b i a double aay, with gavalue in [0, ] (obtained by linea caling).
Image Negative L 0 0 L T ( ) = = L, L : max. gayvalue Film Negative Negative of film negative
Contat Stetching Inceae the dynamic ange of gayvalue in the input image. Suppoe you ae inteeted in tetching the input intenity value in the inteval [, 2 ]: 2 0 2 0 Note that ( - 2 ) < ( - 2 ). The gayvalue in the ange [, 2 ] i tetched into the ange [, 2 ]. Special cae: Theholding o binaization = 2, = 0 and 2 =
2 = = 0 0 = 2 Ueful when we ae only inteeted in the hape of the object and on on thei actual gayvalue. Gaycale Blood cell image a = imead( image.tif ) Binay Blood cell image b = im2bw(a,0.5)
Gamma coection: = 0, 2 =, and 0, < T() = 2 g, 2, > 2 2 = g > 2 = g < = 0 0 2 = 0 0 2 Output Image i dake Output Image i bighte
Example: Gaycale Photogaph a = imead( image.tif ) Contat enhanced photogaph b=imadjut(a,[0.3,0.7], [0,],0.7)
Compeion of Dynamic Range When the dynamic ange of the input gayvalue i lage compaed to that of the diplay, we need to compe the gayvalue ange --- example: Fouie tanfom magnitude. Typically we ue a log cale. = T ( ) = clog( + ) Satun Image Mag. Spectum Mag. Spectum in log cale
Gaylevel Slicing: Highlight a pecific ange of gayvalue. Without backgound With backgound Example: Oiginal Image Highlighted Image (no backgound) Highlighted Image (with backgound)
Bitplane Slicing: Diplay the diffeent bit a individual binay image. 8 bitplane of cameaman image
Image Subtaction In thi cae, the diffeence between two imila image i computed to highlight o enhance the diffeence between them: g( m, = f( m, f2( m, It ha application in image egmentation and enhancement Example: Mak mode adiogaphy f (m, : Image befoe dye injection f 2 (m, : Image afte dye injection g(m, : Image afte dye injection, followed by ubtaction
Image Aveaging fo noie eduction Noie i any andom (unpedictable) phenomenon that contaminate an image. Noie i inheent in mot pactical ytem: Image acquiition Image tanmiion Image ecoding Noie i typically modeled a an additive poce: g( m, = f ( m, + η( m, Noiy Image Noie-fee Image Noie The noie η (m, at each pixel (m, i modeled a a andom vaiable. Uually, η (m, ha zeo-mean and the noie value at diffeent pixel ae uncoelated. Suppoe we have M obevation {g i (m, }, i=, 2,, M, we can (patially) mitigate the effect of noie by aveaging g( m, = M M i= g i ( m,
In thi cae, we can how that E[ g( m, ] = f ( m, Va[ g( m, ] = Va[ η( m, ] M Theefoe, a the numbe of obevation inceae ( M ), the effect of noie tend to zeo.
Image Aveaging Example Noie-fee Image Noiy Image Noie Vaiance = 0.05 M =2 M =5 M =0 M =25 M =50 M =00