Digital Image Processing in Photogrammetry
overview
Part 1 Digital Image Fundamentals Digital Image Quantization, bit depth, histogram Resampling Part 2 Digital Image Processing Fundamentals Smoothing operators Gradient operators Feature extraction Pansharpening
Photogrammetric processes
System level tasks:
Algorithms for storing, accessing, displaying images (e.g. zoom in and out, panning). Manipulation of digital imagery
Low level tasks:
Algorithms for processing and matching images, extract features from images Image processing, orientation, orthophotos, DEM
Middle level tasks:
Algorithms for image segmentation Surface and feature reconstruction
High level tasks:
Algorithms for image understanding Object Recognition, image interpretation
Digital Image
The continuous range of details, intensities and colours are converted to discrete digital values when the image is stored in a computer.
Digital Image
A digital image is a function f(x,y) discretised on the locations (x,y) (i.e. Sampling) and on the brightness levels (i.e. Quantisation) Matrix with x,y as rows and columns and f as matrix element Matrix element = picture element (PIXEL)
Sampling
Quantization
Digital Image
Image = Matrix (nx, ny) =
Digital Image
Discretisation representation of digital images
Quantization B/W Image
Quantization Colour Image
Bit depth
-The human eye can only discern about 10 million different colours -Saving an image in more than 24 bpp is excessive if only for viewing. -Colour gradations in 8bit (or less) images are seen in the image histogram.
Histogram B/W
Histogram: Grey value frequencies
Mean: Reflects the overall brightness of an image Std. Dev. : Measure for the contrast
Histogram - Colour
Histogram: Grey value frequencies
Mean: Reflects the overall brightness of an image Std. Dev. : Measure for the contrast
Histogram manipulation
Histogram: Brightness and Contrast
Histogram manipulation
Histogram Equalization
Resampling
Different operation with digital images require an interpolation of the grey values (e.g. rotation, magnification, etc)
Grey-values in g2(x,y) after the operation are no more in an integer position => Solution: resampling Mainly 4 procedures: (a) Nearest Neighbour (b) Bilinear Interpolation (c) Bicubic Interpolation (d) Spline Applications: Image measurements ia matching Orthophoto generation Image rectification
Resampling
Example
Resampling - Nearest neighbor
- It uses the value of the next pixel as new grey-value - No new grey-value is computed - Fast but not good results - The image intensity is preserved - Histogram before/after resampling is very similar
Resampling Bilinear interpolation
Resampling Bicubic interpolation & Spline
Bicubic -4x4 neighborhood of known pixels (16 pix.) -Closer pixels with higher weighting -Sharper images -Standard in many image editing programs (incl. Adobe Photoshop), printer drivers and incamera interpolation.
Spline
-Computationally expensive -Retain the most image information after an interpolation. -Useful when multiple rotations / distortions are applied. -Not optimal for single step enlargements
Smoothing operators
- Remove noise - Reducing resolution Common convolution filters: - Average - Gaussian - Edge Preserving Smoothing
Smoothing operators
- Remove noise - Reducing resolution Common convolution filters: - Average - Gaussian - Edge Preserving Smoothing (EPS)
Smoothing operators
Gaussian Intensity graph of a step function
smoothed with a Gaussian filter
Smoothing operators : Noise Removal
Original
EPS
Smoothing operators : Noise Removal
Original EPS
Original
EPS
Smoothing operators : Noise Removal
Original Gaussian EPS
Smoothing operators : Reducing Resolution
Gradient operators
- Detect changes in images, occurring within small regions - Used for e.g. edge enhancement, corner detection, detection of road lines, Discontinuities - e.g. Sobel
Feature extraction: Edges
Edges are pixels where the brightness function changes abruptly. Detecting edge pixels : determining local discontinuities in gray levels. Linking edge pixels into edges: identification of edge pixels that belong to the same edge. Grouping edges: identifying straight line segments, polylines, etc.
Feature extraction: Edges
Canny extractor 1. Smoothing: using a gaussian smoothing operator 2. Gradient 3. Non-maximum suppression 4. Hysteresis threshold
Feature extraction: Edges
Canny Grad. Thresh
Feature extraction: Edges
Pansharpening - Methods
A pixel level fusion technique Increase spatial resolution Preserve spectral information in the multispectral data. Different methods: a) Intensity Hue Saturation, b) Principal components analysis, c) Wavelet transformation
Pansharpening
IHS transformation
RGB color space is hardware-oriented: it reflects the use of CRTs. In the HIS color is described by a single representative of one monochromic wave of light Hue with the specification of Saturation measure and Intensity measure.
Pansharpening - example
IKONOS PAN: 1m MS: 4 m
Pansharpening - example
PAN: 0.6 m MS: 2.5 m
Pansharpening - example
Pansharpening - considerations
Factors to be considered before performing sharpening : The application for which the sharpened data is to be used Co-registration of the multispectral and panchromatic images (incl. the geometric corrections and models) Viewing angle of the imagery Time difference in acquisition of panchromatic & multispectral images Resampling algorithm Radiometric quality and preprocessing
