Lecture 4?

Slide 69

  • We measure reflectance but represent them as DN (b). They are related using linear equations to add the gain and offset.

  • Reflectance is %, DN is unitless

  • Irradiance = Reflectance Radiance (R() E())


Page 72

  • In matching, we want to make them look similar. You want the map to appear seamless by possibly using equalization.


Page 75

  • Stretch the dynamic range of values from one limited range to another.

  • You want the range to be more dynamic, while preserving the manipulation you did


Linear Contrast enhancement (p. 76)

  • Ratio and proportion

  • When you have DN range and you want to stretch it, get the DNtarget = DNorig * ((DNtargetmax - DNtargetmin) / (DNorigmax - DNorigmin))


p. 79

  • Use Gaussian Contrast Enhance if the range are very low

  • Instead of using slope equation, we use the gaussian values of the stretch


p.80

  • Segment the original input brightness and map it out to your preffered brightness to make the low brightness higher and vice versa

  • We use interactive to monitor each slope


p. 83

  • Density slicing is done by dividing a series of classifications based on ranges of brightness value

  • You assign a certain value to a pixel if it is within a certain range, and assign corresponding colors for graphical representation

  • This can be done as masking

  • Helpful if there is more than one segment you want to remove


Spatial Enhancement Technique

  • You look how your image values are distributed all throughout your image

  • When there is a variation across the image, there is low spatial frequency. If jumpy, it is high spatial frequency


p. 87

  • Correlations and convolution are operators to represent images using a function


p. 88

  • Use convolution operator to convert only a portion of the image using a kernel window.


p. 91

  • Implemented using common mathematical properties


p. 94

  • Neighborhood is determined by relative size called a window.


p. 96

  • Operate or apply a kernal from one location to the next

  • Can be done on-the-fly nowadays


p. 99

  • Geometric enhancement in terms of kernel window


p. 101

  • Reassignment of image is dependent on some value in terms of the value with a gain and offset

  • Instead of doing it independent of the location, you are looking at the relative location of the pixels itself


p. 103

  • Highpass filters does not preserve the radiometric quality of your image

  • Lowpass filter is the average of pixels by convoluting your image and replacing them

  • Median filter

    • replace with something that exists within the image, no exotic values

    • may possibility matanggal yung pangit na values

  • Logical averaging - it is not necessary to apply manipulation on a square


p. 119

  • You apply a factor (k), HB is the window, f(i,j) is the original window.

  • There is a separation

p. 120

  • If you want to highlight separability

p. 130

  • Subtractive smoothing

  • If you want to have a better contrast between your features. Apply this to the original in one band then make another composite using the transformed band and two others.

  • It will look like it has a high contrast

p. 131

  • Enhance edges

  • Compute the differences in the diagonal direction

p. 135

  • Roberts - nonconforming to fundamental directions

p. 146

  • Tonal differences are preserved in between. Those with gradient becomes flat

p. 149

  • You can apply a system of kernels, changing the box, the values to enhance edges or reduce level of detail