Cellular Automata for Image Resizing - Cellular Automata in Image Processing and Geometry - page 35

Ta b l e 2 . 1 Resulted PSNRS (dB) values. NN: Nearest; BL: Bilinear; BC:Bicubic; ND: NEDI;
EO: Edge - oriented; CA-R method

Image Method

NN BL BC ND EO CA-R

Koala (RGB: 256 × 192) 23.70 25.37 25.14 32.45 29.30 30.32

Cameraman (GR:128 × 128) 22.37 23.96 23.70 30.77 25.80 26.54

Lena (RGB:150 × 150) 26.93 28.88 28.77 34.57 30.90 31.57

Box (GR:320 × 240) 28.68 30.19 29.99 32.1 29.13 30.02

Building (RGB:640 × 480) 29.92 31.71 31.73 33.91 30.62 31.58

Teddy (RGB:450 × 376) 10.9374 11.0288 10.96 16.356 13.3263 15.137

Statue (RGB:384 × 288) 24.9064 26.4758 26.2662 29.167 27.3648 28.473

Butterfly (RGB:133 × 100) 18.5155 20.1773 20.1355 25.1367 23.164 24.526

Port (RGB:800 × 600) 24.4053 25.9650 25.8321 29.3648 27.816 28.7912

Garden (RGB:352 × 240) 16.7522 18.0961 19.7002 23.9463 20.1834 21.648

where MSE denotes the Mean-Squared-Error and is calculated by:

M 1

i = 0

− N 1

i = 0 [ I ( i , j ) − F ( i , j )]

−

1

2

MSE = (2.9)

M × N

where I is the approximated version

of the image and M , N are the dimensions of the image, respectively. For color
images, the definition of PSNR is the exact same except the MSE is the sum over all
squared value differences divided by image size and by three.
All resulted PSNR values for every tested image are provided in Tab. 2.1; while
in Tab. 2.2, the resulted execution time for each method is demonstrated. As it is
depicted in Tab. 2.2, the proposed method produces sufficiently high PSNR values.
Despite their low processing time, the Nearest neighbor method produces zigzag
artifacts over the edge areas while the Bilinear method results in blurred edges. The
Bicubic method also produces blurred edges requiring more processing time. The
highest PSNR values are produced by the NEDI algorithm nevertheless; the required
execution time prohibits its use for real-time applications. The edge-oriented method
produces adequate PSNR values with low processing time however; it is a threshold
dependent method since it is required to determine the edge areas. On the contrary,
the proposed method produces sufficiently high PSNR values preserving the edges
of the initial image. Moreover, exploiting the inherent parallelism of the CA, the
method displays low execution time making appropriate for real time systems.

( i , j ) corresponds to the original image, F ( i , j )