Remote Sensing and Image Processing: 2
Dr. Hassan J. Eghbali
Image processing: image display and enhancement
Purpose• visual enhancement to aid interpretation • enhancement for improvement of information
extraction techniques • Today we will look at image display and
histogram manipulation (contrast ehancement etc.)
Dr. Hassan J. Eghbali
Image Display
• The quality of image display depends on the quality of the display device used– and the way it is set up / used …
• computer screen - RGB colour guns– e.g. 24 bit screen (16777216)
• 8 bits/colour (28)
• or address differently
Dr. Hassan J. Eghbali
Colour Composites‘Real Colour’ compositered band on red
green band on green
blue band on blue
Swanley, Landsat TM
1988
Dr. Hassan J. Eghbali
Colour Composites‘Real Colour’ compositered band on red
Dr. Hassan J. Eghbali
Colour Composites‘Real Colour’ compositered band on red
green band on green
Dr. Hassan J. Eghbali
Colour Composites‘Real Colour’ compositered band on red
green band on green
blue band on blue
approximation to ‘real colour’...
Dr. Hassan J. Eghbali
Colour Composites‘False Colour’ compositeNIR band on red
red band on green
green band on blue
Dr. Hassan J. Eghbali
Colour Composites‘False Colour’ compositeNIR band on red
red band on green
green band on blue
Dr. Hassan J. Eghbali
Colour Composites‘False Colour’ composite• many channel data, much not comparable to RGB (visible)
– e.g. Multi-temporal data
– AVHRR MVC 1995
April
August
September
April; August; September
Dr. Hassan J. Eghbali
Greyscale DisplayPut same information on R,G,B:
August 1995
August 1995
August 1995
Dr. Hassan J. Eghbali
Density Slicing
Dr. Hassan J. Eghbali
Density Slicing
Dr. Hassan J. Eghbali
Density SlicingDon’t always want to use full
dynamic range of display
Density slicing:
• a crude form of classification
Dr. Hassan J. Eghbali
Density SlicingOr use single cutoff
= Thresholding
Dr. Hassan J. Eghbali
Density SlicingOr use single cutoff with
grey level after that point
‘Semi-Thresholding’
Dr. Hassan J. Eghbali
Pseudocolour• use colour to enhance
features in a single band – each DN assigned a
different 'colour' in the image display
Dr. Hassan J. Eghbali
Pseudocolour
• Or combine with density slicing / thresholding
Dr. Hassan J. Eghbali
Histogram Manipluation
• WHAT IS A HISTOGRAM?
Dr. Hassan J. Eghbali
Histogram Manipluation
• WHAT IS A HISTOGRAM?
Dr. Hassan J. Eghbali
Histogram Manipluation
• WHAT IS A HISTOGRAM?
Frequency of occurrence (of specific DN)
Dr. Hassan J. Eghbali
Histogram Manipluation
• Analysis of histogram – information on the dynamic range and
distribution of DN• attempts at visual enhancement
• also useful for analysis, e.g. when a multimodal distibution is observed
Dr. Hassan J. Eghbali
Histogram Manipluation
• Analysis of histogram – information on the dynamic range and
distribution of DN• attempts at visual enhancement
• also useful for analysis, e.g. when a multimodal distibution is observed
Dr. Hassan J. Eghbali
Histogram ManipluationTypical histogram manipulation algorithms:
Linear Transformation
input
outp
ut
0 255
255
0
Histogram ManipluationTypical histogram manipulation algorithms:
Linear Transformation
input
outp
ut
0 255
255
0
Dr. Hassan J. Eghbali
Histogram ManipluationTypical histogram manipulation algorithms:
Linear Transformation
• Can automatically scale between upper and lower limits•or apply manual limits
•or apply piecewise operator
But automatic not always useful ...
Dr. Hassan J. Eghbali
Histogram ManipluationTypical histogram manipulation algorithms:
Histogram EqualisationAttempt is made to ‘equalise’ the frequency distribution across the full DN range
Dr. Hassan J. Eghbali
Histogram ManipluationTypical histogram manipulation algorithms:
Histogram Equalisation
Attempt to split the histogram into ‘equal areas’
Dr. Hassan J. Eghbali
Histogram ManipluationTypical histogram manipulation algorithms:
Histogram Equalisation
Resultant histogram uses DN range in proportion to frequency of occurrence
Dr. Hassan J. Eghbali
Histogram ManipluationTypical histogram manipulation algorithms:
Histogram Equalisation
• Useful ‘automatic’ operation, attempting to produce ‘flat’ histogram
• Doesn’t suffer from ‘tail’ problems of linear transformation
• Like all these transforms, not always successful
• Histogram Normalisation is similar idea
• Attempts to produce ‘normal’ distribution in output histogram
• both useful when a distribution is very skewed or multimodal skewed
Dr. Hassan J. Eghbali
Colour Spaces• Define ‘colour space’ in terms of RGB
• Only for visible part of spectrum:
Dr. Hassan J. Eghbali
Colour Spaces• RGB axes:
Dr. Hassan J. Eghbali
Colour Spaces• RGB (primaries) as axes
Dr. Hassan J. Eghbali
Colour Spaces• Alternative: CMYK ‘subtractive primaries’
• often used for printing (& some TV)
Dr. Hassan J. Eghbali
Colour Spaces• Alternative: CMYK ‘subtractive primaries’
Dr. Hassan J. Eghbali
Colour Spaces• Other important concept: HSI transforms
• Hue (which shade of color)
• Saturation (how much color)
• Intensity
• also, HSV (value), HSL (lightness)
Dr. Hassan J. Eghbali
Colour Spaces• Other important concept: HSI transforms
Dr. Hassan J. Eghbali
Practical 1
• Histogram manipulation – Contrast stretching and histogram equalisation
• visual (qualitative) enhancement of images according to brightness
• Highlight clouds / sea / land in AVHRR image of UK
• Multispectral information – Different DN (digital number) values in different
bands– Due to different properties of surfaces
Dr. Hassan J. Eghbali