Morphological Descriptors and Spatial Aggregations for Characterizing Damaged Buildings in Very High

1
Morphological Descriptors and Spatial
Aggregations for Characterizing Damaged Buildings
in Very High Resolution Images
L. Gueguen, M. Pesaresi, P. Soille and A.
Gerhardinger Geo-Spatial Information Analysis fo
Global Security (ISFEREA) Institute for the
Protection and Security of the Citizen (IPSC)
Joint Research Centre, European
Commission webpage http//isferea.jrc.ec.europa.e
u/
2
Outline

• Context
• Aggregation of morphological descriptors
• Morphological filtering top-hat
• Spatial and fuzzy aggregation
• Damage assessment
• Driving image analysts attention
• Assessment of damaged buildings ratio

3
Context

• Quickbird image
• Pan of size 20433x14933
• MS of size 4496x3287
• Characterize damaged buildings in the image
• damaged buildings have wall and rooms visible in
  the image

4
Aggregation of morphological descriptors
5
Top-hat filtering and fuzzy memberships

• Top-hat filtering by opening or closing are
  morphological based operators
• Top-hat by opening by a structuring element (SE)
  of size s, highlights all bright structures
  smaller than s
• Top-hat by closing by a (SE) of size s,
  highlights all dark structures smaller than s
• These filters enable to characterize walls and
  shadowed rooms of destroyed building

6
Top-hat filetring

• Top-hat by closing by SE of size 4.6x4.6 m2
• Shadowed room descriptor

• Panchromatic image (0.6m resolution)

• Top-hat by opening by SE of size 2x2m2
• Walls descriptor

7
Fuzzy memberships (stretching done visually)
Shadowed room TH by closing linearly stretched
between 0 and 1 Fuzzy membership image f1
Walls TH by opening linearly stretched between 0
and 1 Fuzzy membership image f2

• Panchromatic image (0.6m resolution)

7
8
Spatial aggregation of descriptors (1/2)

• The spatial aggregation should enable to answer
  the question
• How much of the descriptor is in the region W?
• Let f(x) be a morphological descriptor,
  characterizing the pixel x. In other words, f(x)
  represents the degree of membership of x to the
  type of structure.
• The spatial aggregation of f computes the covered
  fuzzy area in W

9
Spatial aggregation of descriptors (2/2)

• Building are of expected size of 15x15 m2
• We implement the spatial aggregation of each
  descriptor f with a low-pass filtering. Taking a
  sliding window W of size 15x15 m2
• Let fW(x), be the spatial aggregation of the
  descriptor f.
• fW(x) is a fuzzy membership indicating the amount
  of descriptor f in region W centered at pixel x.

10
Fuzzy aggregation (1/2)

• Fuzzy logic is used to combine multiple
  descriptors
• Let f1 and f2 be two membership images the fuzzy
  logic fundamental combinations are
• We combine 3 descriptors with AND operator
• Spatial aggregation by W of walls f1W
• Spatial aggregation by W of shadowed rooms f2W
• (1-NDVI) representing non vegetation

11
Descriptors fuzzy aggregation (2/2)
f3
f1W
f2W

• The final result describes
• Structures of 15x15 m2 sizeW
• Structures containing dark structures smaller
  than 5x5m2 f1W
• Structures containing bright structures smaller
  than 2x2m2 f2W
• Structures containing no vegetation f3

AND
result
12
Overlay of results with original image
13
Damage assessment

• Result image helps for image interpretation
• draw image analysts attention to damaged parts
• optimize image analysis time
• Automatic quantitative measurements
• Area and ratio of damaged buildings per region of
  interest

14
Drawing image analysts attention

• Full image (20433x14937)

• Result low pass filtered and colored

15
Assessing the ratio of destroyed building (1/2)

• built-up membership map is extracted (Pantex)
• Texture characteristics
• Spatial and fuzzy based aggregation
• built-up and damaged memberships are combined
  with AND operatorgt damage in built-up
• from built-up and damage in built-up memberships,
  damage ratio are estimated.

damaged buildings
undamaged buildings
16
Assessing the ratio of destroyed building (2/2)

• With administrative region boundaries, the ratio
  of destroyed buildings per region is estimated
• Fuzzy area per regions R is computed for
• Damage in built-up membership D(R)
• Built-up membership B(R)
• The ratio of destroyed building is D(R)/B(R)

1 of destroyed buildings
50 of destroyed buildings
17
Conclusion

• A processing chain for characterizing damaged
  building in VHR images
• Based on morphological descriptors
• Based on spatial and fuzzy aggregation
• Usability of result membership function to damage
  assessment
• To draw image analysts attention
• To assess number or ratio of destroyed building
• Need for procedures automatically adapting to the
  damage type