ECE 468 / CS 519:

Digital Image Processing

Interpolation, Intensity Transforms

Prof. Sinisa Todorovic

[email protected]
Spatial Image Transformations

Affine transforms:

• Translation

• Scaling

• Rotation

• Shear
 Example

(x , y ) = T {(x, y)}
⇥ ⇥
x x
⇤ y ⌅=T⇤ y ⌅ homogeneous
coordinates
1 1
 2D Translation

⇥
x
=
y displacement
⇥
tx
t=
ty homogeneous
coordinates
⇤ ⌅
⇥ ⇥ x
x + tx 1 0 tx ⇧
P =P +t= = y ⌃
y + ty 0 1 ty
1
⇥ ⇥ ⇥
x + tx 1 0 tx x
P ⇤ y + ty ⌅ = ⇤ 0 1 ty ⌅⇤ y ⌅
1 0 0 1 1
⇤
⇥
translation matrix
source: S. Savarese
 2D Scaling

⇥
sx x
=
sy y

⇥
x
=
y

⇥ S ⇥ ⇥
sx x sx 0 0 x
⇧ ⌃ ⇧ ⌃⇧ ⌃
⇤ sy y ⌅ = ⇤ 0 sy 0 ⌅⇤ y ⌅
1 0 0 1 1
⇤
⇥

scaling matrix source: S. Savarese

Complex Affine: First Scaling and then Translation

Is the ordering important?

P =S·P
P = (T · S) · P
P =T ·P
⇥ ⇥ ⇥ ⇥
x 1 0 tx sx 0 0 x
⇧ ⌃ ⇧ ⌃⇧ ⌃⇧ ⌃
⇤ y ⌅ = ⇤ 0 1 ty ⌅ ⇤ 0 sy 0 ⌅⇤ y ⌅
1 0 0 1 0 0 1 1
⇥
sx 0 tx
⇧ ⌃ scaling + translation
A=⇤ 0 sy ty ⌅ matrix
0 0 1 source: S. Savarese
 2D Rotation

counter-clockwise
by angle ⇥
cos x sin y
P =
y sin x + cos y
⇥
y x
P =
y
x x

⇥ R ⇥ ⇥
x cos sin 0 x
⇧ ⌃ ⇧ ⌃⇧ ⌃
⇤ y ⌅ = ⇤ sin cos 0 ⌅⇤ y ⌅
1 0 0 1 1
⇤
⇥
rotation matrix
source: S. Savarese
Complex: First Scaling, Then Rotation, Finally Translation

t R S

⇤
⇥
⇤
⇥
⇤
⇥

translation rotation scaling

matrix matrix matrix

⇤ ⌅ ⇤ ⌅
x ⇥ x
⌥ RS t ⌥
⇧ y ⌃= ⇧ y ⌃
0 1
1 1
 Estimating the Spatial Transform

transformed
input
input

inverse estimation
transform error
 MATLAB Example

>>img=imread(image_name); %input image


>>tform = maketform(’affine’,T); % set transform



>>img_out = imtransform(img, tform, interp);

>>imshow(img_out)
 Image Interpolation

original resampling shrinking zooming

Image Interpolation
Bilinear Interpolation

f (x, y) = ax + by + cxy + d

coefficients that need to be estimated

 Bilinear Interpolation

f (x, y) = ax + by + cxy + d

coefficients that need to be estimated

?
 Bilinear Interpolation

f (x, y) = ax + by + cxy + d

coefficients that need to be estimated

nearest neighbor
 Bilinear Interpolation

f (x, y) = ax + by + cxy + d

coefficients that need to be estimated

?
 Bilinear Interpolation

f (x, y) = ax + by + cxy + d

coefficients that need to be estimated

nearest neighbor
 Bilinear Interpolation

f (x, y) = ax + by + cxy + d

1 2 coefficients that need to be estimated

?
4 3

nearest neighbor
 Bilinear Interpolation

f (x, y) = ax + by + cxy + d

1 2 coefficients that need to be estimated

? ax1 + by1 + cx1 y1 + d = f (x1 , y1 )

4 3
ax2 + by2 + cx2 y2 + d = f (x2 , y2 )
ax3 + by3 + cx3 y3 + d = f (x3 , y3 )
ax4 + by4 + cx4 y4 + d = f (x4 , y4 )
nearest neighbor

a, b, c, d
 Example: Zooming-in

nearest neighbor bilinear

 Image Interpolation

• Bilinear N = 1

• Bicubic N = 3

N N
f (x, y) = i j
aij x y
i=0 j=0

new locations new locations

estimated from the known neighboring locations

Intensity Transformations of Images

g(x, y) = T {f (x, y)}

transformed input
input
 Example: Shading Correction

g(x, y) = f (x, y)h(x, y)

input shading profile f(x,y)/shading

 Example: Masking

g(x, y) = f (x, y)h(x, y)

input mask f(x,y) * mask

 Rescaling Intensity Values

f (x, y) min[f (x, y)]

g(x, y) = K
max[f (x, y)] min[f (x, y)]

input output
 Image Negatives

g(x, y) = L 1 f (x, y)
 Contrast Stretching
input

output thresholded
Gamma Correction

s = cr
 Example: Gamma Correction

input γ=.6 input γ=3

γ=.4 γ=.3 γ=4 γ=5