Digital ImageDisplay Basics

1
Digital Image/Display Basics
2
Scale vs. Spatial Resolution

• Scale is best used to describe the quality of
  vector information based on what is the precision
  of the map source.

3
Scale vs. Spatial Resolution

• Scale is best used to describe the quality of
  vector information based on what is the precision
  of the map source.
• Spatial resolution is used to describe the
  quality of the raster information based on the
  minimum ground size of the raster cell.

4
Scale vs. Spatial Resolution

• Scale is best used to describe the quality of
  vector information based on what is the precision
  of the map source.
• Spatial resolution is used to describe the
  quality of the raster information based on the
  minimum ground size of the raster cell.
• These terms are rather mixed together,
  especially with air photos. Air photo projects
  are contracted to be flown at a certain scale.
  But when scanned in the air photos quality is
  best described by its spatial resolution which is
  directly a result of the scanning resolution but
  indirectly also related to its original scale and
  the resolution of the photo grain.

5
Sensor Spatial Resolution

• Spatial resolution of a multi-spectral sensor is
  defined by its Instantaneous Field of View (IFOV)
  and is a function of
• altitude of the sensor,
• optics and sensitivity of the sensor,
• dwell time,
• and amount of energy available in the bandwidth.

6
Spatial Resolution
Spatial resolution of some publicly available
satellite sensors.
7
Radiometric Resolution

• Once detected by the sensor, the response is
  digitally recorded as binary data. The higher
  the response, the higher value recorded (this is
  as an integer which is known as a DN value
  digital number value).
• Binary data records the data in a series of bits
  with each bit switched to on or off (0 or 1).
  The pattern of bit response then is translated by
  a computer as to the DN value.
• The higher the bit value of the system the more
  DN values it can record, the more detail is
  available.

8
Radiometric Resolution

• Once detected by the sensor, the response is
  digitally recorded as binary data. The higher
  the response, the higher value recorded (this is
  as an integer which is known as a DN value
  digital number value).
• Binary data records the data in a series of bits
  with each bit switched to on or off (0 or 1).
  The pattern of bit response then is translated by
  a computer as to the DN value.
• The higher the bit value of the system the more
  DN values it can record, the more detail is
  available.

9
Radiometric Resolution

• Once detected by the sensor, the response is
  digitally recorded as binary data. The higher
  the response, the higher value recorded (this is
  as an integer which is known as a DN value
  digital number value).
• Binary data records the data in a series of bits
  with each bit switched to on or off (0 or 1).
  The pattern of bit response then is translated by
  a computer as to the DN value.
• The higher the bit value of the system the more
  DN values it can record, the more detail is
  available.

10
Radiometric Resolution

• Example A 1-bit system has only 2 responses
• 0 or 0
• 1 or 1
• This can be calculated by
• 21 2

11
Radiometric Resolution

• Example A 2-bit system has 4 responses
• 0,0 or 0
• 0,1 or 1
• 1,0 or 2
• 1,1 or 3
• This can be calculated by
• 22 4

12
Radiometric Resolution

• And on and on . Some of the earlier systems such
  as Landsat MSS was a 7-bit system or 27 (128 DN
  values). Many systems are 8-bit or 28 (256 DN
  values). Some are even 11-bit (2,048 DN values)
  or even 16-bit (65,536 DN values). This range of
  values is called the systems dynamic range and
  defines its radiometric resolution.
• Note Humans can generally perceive only 64
  shades of gray.

13
Radiometric Resolution

• Most photo scanners, scan in a BW photo at 8-bit
  resolution.
• For a color photo, they will usually give two
  options
• 24-bit color Red, green and blue are scanned
  and create a 3-band output image of red, green,
  and blue in 8-bit.
• 8-bit color A color from a 256 value color
  table is assigned from a Look-Up Table
  (pseudo-color mode).
• Also all output display systems tend to be in
  8-bit grayscale or 24-bit RGB color.

14
Radiometric Resolution

• Another way of looking at radiometric resolution
  with a 1-bit gray scale on top grading down
  towards an 8-bit gray scale on the bottom.
• In addition, data can be unsigned (only positive
  values) or signed (negative or positive values).

15
Radiometric Resolution

• Knowing Radiometric Resolution makes you cool
  enough to pull off this T-Shirt.

16
Computer Display

• Computer screens typically display image
  information three ways
• one band (layer) can be displayed in grayscale
  (what could be called black and white).

17
Computer Display

• Computer screens typically display image
  information three ways
• one band (layer) can be displayed in grayscale
  (what could be called black and white).
• one band can have each of its data values
  assigned a color through a Look-Up Table (LUT) in
  what is called pseudo-color.

18
Computer Display

• Computer screens typically display image
  information three ways
• one band (layer) can be displayed in grayscale
  (what could be called black and white).
• one band can have each of its data values
  assigned a color through a Look-Up Table (LUT) in
  what is called pseudo-color.
• up to three bands can be displayed in color
• with one band displayed in the red color display
  gun.
• another band displayed in the green color
  display gun.
• and another displayed in the blue color display
  gun.
• This is known as a False Color Composite (FCC)
  with the combinations of Red, Green and Blue
  (also called a RGB display) able to make all of
  the colors of the rainbow.

19
Computer Display

• Computer screens typically display image
  information three ways
• Continuous (gradient) image data are usually
  displayed in single bands as grayscale or in
  multi-band in FCCs. Continuous data sets include
  raw image bands, DEMs, thermal data.
• Thematic image data sets such as classifications
  are usually displayed as pseudo-color.

20
Grayscale Image

• DN values are displayed as a value of gray
  depending on how large the number. Ex in an
  8-bit system 0 is black and 255 is white with
  everything else a shade of gray.

21
Grayscale Image

• This signed 16-bit DEM of the US is being
  displayed as a grayscale image with its lowest
  value of 79 m below sea level as black, its
  highest value of 4,328 m above sea level as white
  and every value in between as gradient of gray.

22
Pseudo-color Image

• A pseudo-color image can be a continuous or
  thematic gray-scale image in which its values are
  passed through a LUT color table. This can be
  done to create a simple color-ramp - in this case
  a DEM is color-ramped with the cooler colors
  representing low elevations and the hotter colors
  representing higher elevations.

23
Pseudo-color Image

• Or through a simple classification procedure
  known as a density-slice in which data value
  ranges of a gray-scale image are color coded - in
  this case the same DEM is color-coded with deep
  blue representing below sea level, light blue
  representing from 0-7 m above sea level and tan
  representing 8 m and above sea level.

24
Additive Colors RGB Display

• Additive primary colors (Red, Green, and Blue or
  RGB) can be added to create all colors. Equal
  proportions or two primary colors create additive
  complementary colors of Magenta (RB), Yellow
  (RG), and Cyan (GB). Combining all three in
  equal proportions creates gray ranging from black
  (no color) to white (complete saturation)
  depending on intensity.

25
Additive Colors RGB Display

• This is the color property found in surfaces that
  emit ?s such as the sun or computer monitors and
  television screens.

26
Additive Colors RGB Display

• Another way of looking at it a mixture of low
  responses in RGB creates black.

27
Additive Colors RGB Display

• Another way of looking at it a mixture of a
  high response in red, moderate response in green
  and a low responses in blue creates brown.

28
Additive Colors RGB Display

• Another way of looking at it a mixture of a
  high response in blue, moderate response in red
  and a low responses in green creates purple.

29
Subtractive Colors CMY(K)

• Another color system are the subtractive colors
  (Cyan, Magenta, and Yellow or CMY) and added
  together they can create most colors. Combined
  in equal proportions they can create the primary
  additive colors Red (YM), Green (YC), and Blue
  (CM), and combined all together they can create
  gray from white (no color) to black (complete
  saturation).

30
Subtractive Colors CMY(K)

• This is the color system used by surface which
  have reflected ?s such as printed documents or
  film. Even though black can be created by CMY,
  printers often a separate black ink in order to
  save on ink.
• NOTE CMY is a subset of RGB and therefore not
  all of the colors are available in this system.

31
Image Display

• Before being displayed on the computer screen the
  image data goes through an invisible,
  instantaneous process of having its numbers
  change according to a Look-Up Table (LUT).

32
Image Display

• Before being displayed on the computer screen the
  image data goes through an invisible,
  instantaneous process of having its numbers
  change according to a Look-Up Table (LUT).
• This was already mentioned with the Pseudo-color
  option, but it also happens with the other two
  display options. The reason for this is to make
  the image more displayable add more contrast.

33
Image Display

• To understand how this happens you need to
  understand some descriptive univariate statistics
  terms

34
Image Display

• To understand how this happens you need to
  understand some descriptive univariate statistics
  terms
• Measurements of Central Tendency
• Mean (?) The average of the data values.
• Mode The one value most represented in the data
  range.
• Median The value in the middle of a data range.

35
Image Display

• To understand how this happens you need to
  understand some descriptive univariate statistics
  terms
• Measurements of Variance
• Dynamic Range The number of values from the
  maximum value to the minimum value.
• Variance (s2) The sum of the difference between
  each value and the mean squared.
• Standard Deviation (s) The square root of the
  variance.

36
Image Display

• To understand how this happens you need to
  understand some descriptive univariate statistics
  terms
• A natural data set will have a normal
  distribution curve a bell curve.

37
Image Display

• To understand how this happens you need to
  understand some descriptive univariate statistics
  terms
• A natural data set will have a normal
  distribution curve a bell curve.

In a perfect normal distribution, the median, the
mode and the mean would equal each other. This is
also called a unimodal distribution.
38
Image Display

• To understand how this happens you need to
  understand some descriptive univariate statistics
  terms
• Most image data is not perfectly normally
  distributed, as in the histogram below, they can
  be multimodal and skewed.

Mode
In this case, the median, the mode and the mean
are not equal to each other. This is a multimodal
distribution skewed to the right (tail to the
right and the median and mode are less than the
mean the opposite is true with a left-skewed
distribution).
Median
Mean
39
Image Display

• To maximize the image data dynamic range to the
  full display capability of the computer screen,
  the image is contrast stretched or enhanced.
  Most contrast enhancement techniques assume
  relative normality in the data and use a linear
  stretch. Examples are
• Simple Linear Input data values are equal to
  output display values.

40
Image Display

• To maximize the image data dynamic range to the
  full display capability of the computer screen,
  the image is contrast stretched or enhanced.
  Most contrast enhancement techniques assume
  relative normality in the data and use a linear
  stretch. Examples are
• Simple Linear Input data values are equal to
  output display values.
• Min/Max Cutoff Output display values are set to
  maximize range between the minimum and maximum
  values.

41
Image Display

• To maximize the image data dynamic range to the
  full display capability of the computer screen,
  the image is contrast stretched or enhanced.
  Most contrast enhancement techniques assume
  relative normality in the data and use a linear
  stretch. Examples are
• Simple Linear Input data values are equal to
  output display values.
• Min/Max Cutoff Output display values are set to
  maximize range between the minimum and maximum
  values.
• Standard Deviation Similar to Min/Max, but the
  minimum and maximum values are set as a /- s
  distance from the ?. This is Erdass default
  stretch which is 2 ss from the ? (Under normal
  conditions this would stretch 95 of the data).

42
Image Display

• To maximize the image data dynamic range to the
  full display capability of the computer screen,
  the image is contrast stretched or enhanced.
  Sometimes the image are far from being normal and
  require non-linear techniques. Examples are
• Piecewise Linear Input data values have
  different output display ranges.

43
Image Display

• To maximize the image data dynamic range to the
  full display capability of the computer screen,
  the image is contrast stretched or enhanced.
  Sometimes the image are far from being normal and
  require non-linear techniques. Examples are
• Piecewise Linear Input data values have
  different output display ranges.
• Histogram Equalization Sorts the input
  histogram proportionally to create as normal an
  output histogram as possible.

44
Image Display

• To maximize the image data dynamic range to the
  full display capability of the computer screen,
  the image is contrast stretched or enhanced.
  Sometimes the image are far from being normal and
  require non-linear techniques. Examples are
• Piecewise Linear Input data values have
  different output display ranges.
• Histogram Equalization Resorts the input
  histogram proportionally to create as normal an
  output histogram as possible.
• Gamma It imposes an exponential stretch for
  data that is tremendously skewed.

45
Image Display

• Some Examples of this are shown above in which
  the histogram on top goes through no stretch, a
  linear stretch (like a min-max cutoff), a
  histogram equalization, and a special stretch
  (such as a gamma) from Lillesand Kiefer,
  Remote Sensing and Image Interpretation, 4th
  ed..

46
Image Display

• To maximize the image data dynamic range to the
  full display capability of the computer screen,
  the image is contrast stretched or enhanced.
• Remember, image contrast enhancement only
  operates on the LUT display table, it does
  nothing to the actual image data values.

47
Other Image Notes

• Images are raster data sets or grids made up by a
  series of cells (grids or pixels). Each cell
  represents the smallest bit of information that
  can be retrieved from the image this defines
  its spatial resolution.

48
Other Image Notes

• Images are raster data sets or grids made up by a
  series of cells (grids or pixels). Each cell
  represents the smallest bit of information that
  can be retrieved from the image this defines
  its spatial resolution.
• The cells along the x-axis are called columns
  and along the y-axis are called rows.

49
Other Image Notes

• Images are raster data sets or grids made up by a
  series of cells (grids or pixels). Each cell
  represents the smallest bit of information that
  can be retrieved from the image this defines
  its spatial resolution.
• The cells along the x-axis are called columns
  and along the y-axis are called rows.
• The image will geo-reference itself, by knowing
  the coordinates at the upper-left hand corner and
  then counting the number of columns and rows to a
  cell and multiplying that by the spatial
  resolution of each cell.

50
Other Image Notes

• Images are raster data sets or grids made up by a
  series of cells (grids or pixels). Each cell
  represents the smallest bit of information that
  can be retrieved from the image this defines
  its spatial resolution.
• By knowing the number of columns and rows of an
  image (or calculating it based on the spatial
  resolution of the cells and the scene size), the
  disk space used by the image file can be
  calculated by
• ( of columns) ( of rows) ( of bands) for
  8-bit data

51
Other Image Notes

• Images are raster data sets or grids made up by a
  series of cells (grids or pixels). Each cell
  represents the smallest bit of information that
  can be retrieved from the image this defines
  its spatial resolution.
• By knowing the number of columns and rows of an
  image (or calculating it based on the spatial
  resolution of the cells and the scene size), the
  disk space used by the image file can be
  calculated by
• ( of columns) ( of rows) ( of bands) for
  8-bit data
• ( of columns) ( of rows) ( of bands) 2
  for 16-bit data

52
Other Image Notes

• Images are raster data sets or grids made up by a
  series of cells (grids or pixels). Each cell
  represents the smallest bit of information that
  can be retrieved from the image this defines
  its spatial resolution.
• By knowing the number of columns and rows of an
  image (or calculating it based on the spatial
  resolution of the cells and the scene size), the
  disk space used by the image file can be
  calculated by
• ( of columns) ( of rows) ( of bands) for
  8-bit data
• ( of columns) ( of rows) ( of bands) 2
  for 16-bit data
• ( of columns) ( of rows) ( of bands) 4
  for 32-bit or floating point data

53
Other Image Notes

• ERDAS also adds one more calculus to this. In
  addition to the .IMG file it creates for the
  actual image data, it creates pyramid layers or
  Reduced Raster Datasets (.RRDs). These files are
  reduced resamples of the images files 1 cell
  for every 2x2 window, then 4x4 window and on and
  on. These .RRD files are used when zooming out
  of the image to save the time of resampling every
  pixel on the screen which can take considerable
  time. These files add another third again to the
  overall file size.