Next: Bibliography
Up: Texture Analysis and Synthesis
Previous: More Accurate First Approximation
Synthesised Natural Textures
The textures presented in
Figs B.1- B.12
were synthesised with the bunch sampling algorithm outlined in
Chapter. 7. The input and the synthesised images are
of size
and
respectively. The
window size for recovering an MBIM is
, so each MBIM
is of size
. The input textures are mainly taken
from [8,82,65].
These synthesised textures confirm that the bunch sampling and the
underlying generic MGRF model capture and replicate the global
periodicity structures in the training images very effectively. This
can be shown by synthesised homogeneous textures, such as `D1' and
`D11' in Fig B.1, `Floral5' and
`Cloth37' in Fig B.7, `S30' and
`Design1' in Fig B.10, and `Blue
Ribbon', `Check-board' and "Manhole Cover' in
Fig B.11, in which repetitive patterns
in the input textures are successfully replicated. This can also be
confirmed by synthesised (rectified) weakly-homogenous textures,
such as `D53' in Fig B.1, 'Fabric0000'
in Fig B.4, `Fabric0013' in
Fig B.5, `S9' in
Fig B.10, and `Campbell' and `Fish
Fabric' in Fig B.11, in which the
global periodicity structures are still preserved although local
details are more or less lost during the process.
The synthesised textures show that the bunch sampling fails on
inhomogeneous images such as `D66' and `D74' in
Fig B.2, `Build0008' in
Fig B.4, and `No22' and `Mountain' in
Fig B.9.
In addition, the synthesis fails if the search window is not
adequately large to accommodate the repetition structure, which is
demonstrated by synthesised textures like `D94' in
Fig B.3 and `CircleBlob' in
Fig B.10.
In synthesising stochastic textures, the bunch sampling produces
general better results for textures with shorter pixel interactions.
These textures include `D29' in
Fig B.1, `Food0005' and 'Misc0003' in
Fig B.6, `Fros' and `Img0036' in
Fig B.8, and `Rockwall1' in
Fig B.9. For stochastic textures
containing coherent local objects, such as `food0007' and
`leave0010' in Fig B.6, `Colour Beads'
in Fig B.7, `Grass13' and `Leaf1' in
Fig B.8, and `Red Peppers' in
Fig B.12, the bunch sampling tends to
cut objects to average size and might leads to visual artifacts.
Nevertheless, from the results, it can be ascertained that on the
whole the generic MGRF model and the bunch sampling algorithm have
potentiality in modelling and synthesising a wide range of natural
textures.
Figure B.1:
Synthesised textures by bunch
sampling: The sizes of training textures, MBIMs, and synthetic
textures are
,
and
,
respectively. The training textures are taken from [8].
|
|
|
MBIM |
|
|
|
|
D1 |
|
|
|
|
MBIM |
|
|
|
|
D9 |
|
|
|
|
MBIM |
|
|
|
|
D11 |
|
|
|
|
MBIM |
|
|
|
|
D12 |
|
|
|
|
MBIM |
|
|
|
|
D23 |
|
|
|
|
MBIM |
|
|
|
|
D29 |
|
|
|
|
MBIM |
|
|
|
|
D35 |
|
|
|
|
MBIM |
|
|
|
|
D53 |
|
|
Figure B.2:
Synthesised textures by bunch
sampling: The sizes of training textures, MBIMs, and synthetic
textures are
,
and
,
respectively. The training textures are taken from citeBrodatz.
|
|
|
MBIM |
|
|
|
|
D55 |
|
|
|
|
MBIM |
|
|
|
|
D56 |
|
|
|
|
MBIM |
|
|
|
|
D65 |
|
|
|
|
MBIM |
|
|
|
|
D66 |
|
|
|
|
MBIM |
|
|
|
|
D68 |
|
|
|
|
MBIM |
|
|
|
|
D69 |
|
|
|
|
MBIM |
|
|
|
|
D74 |
|
|
|
|
MBIM |
|
|
|
|
D76 |
|
|
Figure B.3:
Synthesised textures by bunch
sampling: The sizes of training textures, MBIMs, and synthetic
textures are
,
and
,
respectively. The training textures are taken from citeBrodatz.
|
|
|
MBIM |
|
|
|
|
D77 |
|
|
|
|
MBIM |
|
|
|
|
D79 |
|
|
|
|
MBIM |
|
|
|
|
D80 |
|
|
|
|
MBIM |
|
|
|
|
D83 |
|
|
|
|
MBIM |
|
|
|
|
D92 |
|
|
|
|
MBIM |
|
|
|
|
D93 |
|
|
|
|
MBIM |
|
|
|
|
D94 |
|
|
|
|
MBIM |
|
|
|
|
D105 |
|
|
Figure B.4:
Synthesised
textures by bunch sampling: The sizes of training textures, MBIMs,
and synthetic textures are
,
and
,
respectively. The training textures are taken from [82].
|
|
|
MBIM |
|
|
|
|
Bark0000 |
|
|
|
|
MBIM |
|
|
|
|
Bark0001 |
|
|
|
|
MBIM |
|
|
|
|
Bark0011 |
|
|
|
|
MBIM |
|
|
|
|
Brick0000 |
|
|
|
|
MBIM |
|
|
|
|
Brick0006 |
|
|
|
|
MBIM |
|
|
|
|
Build0008 |
|
|
|
|
MBIM |
|
|
|
|
Fabric0000 |
|
|
|
|
MBIM |
|
|
|
|
Fabric0002 |
|
|
Figure B.5:
Synthesised
textures by bunch sampling: The sizes of training textures, MBIMs,
and synthetic
textures are
,
and
,
respectively. The training textures are taken from [82].
|
|
|
MBIM |
|
|
|
|
Fabric0004 |
|
|
|
|
MBIM |
|
|
|
|
Fabric0007 |
|
|
|
|
MBIM |
|
|
|
|
Fabric0013 |
|
|
|
|
MBIM |
|
|
|
|
Fabric0015 |
|
|
|
|
MBIM |
|
|
|
|
Flower0001 |
|
|
|
|
MBIM |
|
|
|
|
Flower0003 |
|
|
|
|
MBIM |
|
|
|
|
Flower0005 |
|
|
|
|
MBIM |
|
|
|
|
Food0000 |
|
|
Figure B.6:
Synthesised
textures by bunch sampling: The sizes of training textures, MBIMs,
and synthetic
textures are
,
and
,
respectively. The training textures are taken from [82].
|
|
|
MBIM |
|
|
|
|
Food0003 |
|
|
|
|
MBIM |
|
|
|
|
Food0005 |
|
|
|
|
MBIM |
|
|
|
|
Food0007 |
|
|
|
|
MBIM |
|
|
|
|
Leave0010 |
|
|
|
|
MBIM |
|
|
|
|
Metal0002 |
|
|
|
|
MBIM |
|
|
|
|
Misc0003 |
|
|
|
|
MBIM |
|
|
|
|
Water0000 |
|
|
|
|
MBIM |
|
|
|
|
Water0004 |
|
|
Figure B.7:
Synthesised textures by bunch
sampling: The sizes of training textures, MBIMs, and synthetic
textures are
,
and
,
respectively. The training textures are taken from [65].
|
|
|
MBIM |
|
|
|
|
Nature07 |
|
|
|
|
MBIM |
|
|
|
|
Nature10 |
|
|
|
|
MBIM |
|
|
|
|
Nature16 |
|
|
|
|
MBIM |
|
|
|
|
Nature21 |
|
|
|
|
MBIM |
|
|
|
|
Cassava |
|
|
|
|
MBIM |
|
|
|
|
Cloth37 |
|
|
|
|
MBIM |
|
|
|
|
Colour Beads |
|
|
|
|
MBIM |
|
|
|
|
Floral5 |
|
|
Figure B.8:
Synthesised textures by bunch
sampling: The sizes of training textures, MBIMs, and synthetic
textures are
,
and
,
respectively. The training textures are taken from [65].
|
|
|
MBIM |
|
|
|
|
Foliage2 |
|
|
|
|
MBIM |
|
|
|
|
Fros01 |
|
|
|
|
MBIM |
|
|
|
|
Grass13 |
|
|
|
|
MBIM |
|
|
|
|
Img0036 |
|
|
|
|
MBIM |
|
|
|
|
Leaf1 |
|
|
|
|
MBIM |
|
|
|
|
Leaf2 |
|
|
|
|
MBIM |
|
|
|
|
Marble-Black4 |
|
|
|
|
MBIM |
|
|
|
|
Marble-Blue5 |
|
|
Figure B.9:
Synthesised textures by bunch
sampling: The sizes of training textures, MBIMs, and synthetic
textures are
,
and
,
respectively. The training textures are taken from [65].
|
|
|
MBIM |
|
|
|
|
Marble-Brown5 |
|
|
|
|
MBIM |
|
|
|
|
Mossy Bark |
|
|
|
|
MBIM |
|
|
|
|
Mountain |
|
|
|
|
MBIM |
|
|
|
|
No22 |
|
|
|
|
MBIM |
|
|
|
|
Olives |
|
|
|
|
MBIM |
|
|
|
|
Wood-Lite4 |
|
|
|
|
MBIM |
|
|
|
|
Pie |
|
|
|
|
MBIM |
|
|
|
|
Rockwall1 |
|
|
Figure B.10:
Synthesised textures by bunch
sampling: The sizes of training textures, MBIMs, and synthetic
textures are
,
and
,
respectively. The training textures are taken from [65].
|
|
|
MBIM |
|
|
|
|
S30 |
|
|
|
|
MBIM |
|
|
|
|
S7 |
|
|
|
|
MBIM |
|
|
|
|
S9 |
|
|
|
|
MBIM |
|
|
|
|
Bamboo |
|
|
|
|
MBIM |
|
|
|
|
Flake |
|
|
|
|
MBIM |
|
|
|
|
Circle Blob |
|
|
|
|
MBIM |
|
|
|
|
Design1 |
|
|
|
|
MBIM |
|
|
|
|
Design22 |
|
|
Figure B.11:
Synthesised textures by bunch
sampling: The sizes of training textures, MBIMs, and synthetic
textures are
,
and
,
respectively. The training textures are taken from [65].
|
|
|
MBIM |
|
|
|
|
Campbell |
|
|
|
|
MBIM |
|
|
|
|
Blue Ribbon |
|
|
|
|
MBIM |
|
|
|
|
Zesta |
|
|
|
|
MBIM |
|
|
|
|
Check-board |
|
|
|
|
MBIM |
|
|
|
|
Wire Mesh |
|
|
|
|
MBIM |
|
|
|
|
Fish Fabric |
|
|
|
|
MBIM |
|
|
|
|
Manhole Cover |
|
|
|
|
MBIM |
|
|
|
|
Metal18 |
|
|
Figure B.12:
Synthesised
textures by bunch sampling: The sizes of training textures, MBIMs,
and synthetic textures are
,
and
, respectively. The
training textures are taken from [65].
|
|
|
MBIM |
|
|
|
|
Rock01 |
|
|
|
|
MBIM |
|
|
|
|
Metal04 |
|
|
|
|
MBIM |
|
|
|
|
Plaid |
|
|
|
|
MBIM |
|
|
|
|
Red Peppers |
|
|
|
|
MBIM |
|
|
|
|
Puzzle |
|
|
|
|
MBIM |
|
|
|
|
Blue Pattern |
|
|
|
|
MBIM |
|
|
|
|
Board13 |
|
|
|
|
MBIM |
|
|
|
|
Cracks |
|
|
Next: Bibliography
Up: Texture Analysis and Synthesis
Previous: More Accurate First Approximation
dzho002
2006-02-22