LF2301 LODEV [LOGIC Devices Incorporated], LF2301 Datasheet - Page 9

• Manufacturer Part Number: LF2301
• Description: Image Resampling Sequencer
• Manufacturer: LODEV [LOGIC Devices Incorporated]

DEVICES INCORPORATED
Image Rotation & Bilinear Interpolation
Figure 8 shows an example of rotating
an image 30º and using bilinear
interpolation. This mode is selected by
loading M
spiral walk. A counterclockwise spiral
walk could be selected by loading M
with “10.” Bilinear interpolation
requires a kernel size of 2 x 2 pixels.
Loading K
kernel size of 2 x 2. The first pixel
selected is determined by x
this example, the first pixel is (0,0). In
this case, the LF2301s should address
consecutive pixels during each spiral
walk. For this to occur, FOV must be
set to 1 (F
After the last pixel of a spiral walk has
been selected, the next pixel address is
determined by adding dx/du to the
current X address and by adding
dy/du to the current Y address
(unless the kernel just completed was
the last for that line). At the end of the
first spiral walk, pixel (0,1) is ad-
dressed. Since the next calculated
pixel should be (0.866,0.5), dx/du is
selected to be 0.866 and dy/du is
selected to be 0.5. However, after
adding dx/du and dy/du to the X
and Y addresses respectively, the
generated address is (0.866,1.5). The Y
address is off by a value of 1. This is
due to the fact that the last pixel
address of a spiral walk is used to
calculate the first pixel address of
the next spiral walk. In order for the
LF2301s to generate the correct result,
dy/du must be modified by subtract-
ing a 1 from it. The correct value of
dy/du is -0.5. Figure 6 shows how the
unmodified differential terms were
calculated.
After the last pixel of the last spiral
walk on the first line has been selected,
the first pixel address of the second
line is determined by adding dx/dv to
x
the first calculated pixel of the first
0
and by adding dy/dv to y
2-0
1-0
3-0
loaded with “001”).
with “00” for a clockwise
with “0001” selects a
0
and y
0
. Since
0
. In
1-0
spiral walk on the second line should
be (-0.5,0.866), dx/dv is selected to
be -0.5 and dy/dv is selected to be
0.866. Second order differential terms
are not used in this transform and are
therefore set to 0.
It is important to note that the integer
portion of the address generated in
the LF2301 is used as the X or Y
pixel address. The fractional portion
(sub-pixel portion) is used as the
coefficient RAM address.
UMIN and VMIN are both selected to
be 0. UMAX and VMAX are both
selected to be 2. Table 7 shows the
values loaded into all Parameter
Registers. Table 8 shows the ITS
outputs for this example.
F
F
IGURE
IGURE
dy
dv
du
du
dx
dy
dx
dv
dy
dv
source image pixels
calculated pixels
6. D
7. 30 I
= cos 30º
=
=
=
dx
dv
sin 30º = 0.5
–sin 30º = –0.5
cos 30º = 0.866
y
IFFERENTIAL
2-9
MAGE
dx
du
30º
= 0.866
R
OTATION
T
ERMS
dy
du
Image Resampling Sequencer
x
ADDR Row (HEX) Column (HEX)
F
T
0
1
2
3
4
0000
0001
0010
0011
0100
0101
0110
0111
1000
1001
1010
1011
1100
1101
1110
1111
IGURE
1 = 1st pixel of 1st spiral walk,
2 = 1st pixel of 2nd spiral walk, etc.
ABLE
Video Imaging Products
-1
source image pixels
calculated pixels
5
7. P
8. 30 I
0
DDB
1,2
FFF
1FF
000
000
000
100
800
000
000
000
000
000
000
000
002
ARAMETER
MAGE
1
3
08/16/2000–LDS.2301-H
R
R
2
LF2301
OTATION
DDB
EGISTERS
FFF
1FF
000
000
100
800
100
000
000
000
000
000
000
000
002
4
3