Section 2 Image quality, radiometric analysis, preprocessing Emmanuel Baltsavias
Radiometric Quality (refers mostly to Ikonos) Preprocessing by Space Imaging (similar by other firms too): Modulation Transfer Function Correction (MTFC) Always performed Sharpen image especially in scan direction due to TDI imaging (typically 13 lines) or due to satellite rotation during imaging of one line, which cause blurring Dynamic Range Adjustment (DRA) Performed optionally Stretch grey values to better occupy grey value range Some artifacts are due to compression from 11 to 2.6 bit (visible esp. in homogeneous areas) With some sensors compression up to factor 9 (ALOS) or 10 (Resurs-DK-1) are applied!
Radiometric Quality 11bit histogram Nadir PAN (Melbourne) - without DRA 0 512 2048 11bit histogram Nadir PAN (Luzern) - with DRA 0 512 2048 D R A stretches the grey values (GVs) to cover more uniformly the 11 bit range. Result : Absolute radiometric accuracy is destroyed + leads to combination of GVs that are not frequently occupied. Better methods of contrast stretch exist. Suggestion: order images with DRA only for 8-bit images and visual (manual) processing.
Radiometric Quality Important aspects for Feature Extraction and Interpretation Pan-Sharpened 1m Ikonos (7 0 tilt, summer end) Stereo 1m Ikonos (29 0 tilt, winter) View angle Sun angle & Shadowing Season Atmospheric conditions Stereo or mono Colour or B&W Image preprocessing factors over which there is no or limited user control
Radiometric Quality Image quality / interpretability can vary dramatically Images taken the same day of April from the same orbit Luzern (CH) Greek village (Nisyros)
Radiometric Quality Role of shadows and saturation (bright walls)
Image feature variation - Ikonos GEO 1m pan sharpened (RGB), Chinese military base in Hainan Similar sun elevation / azimuth, quite similar sensor elevation 4 / 4 / 2001 9 / 4 / 2001 10 / 4 / 2001 30 / 4 / 2001
Radiometric Quality Noise characteristics analyzed in areas: homogeneous (lake and sea surfaces) Image type Mean std. dev. PAN-MSI 5.2 MSI 2.0 PAN 4.6 PAN-DRA 5.0 Noise generally high since 11bit data represent 8-9 effective bits
Radiometric Quality Noise characteristics analyzed in areas (PAN images): non-homogeneous (whole image excluding large homog. areas) GV range 0-127 128-255 256-383 384-511 512-639 640-767 Raw Image 2.6 3.1 4.1 4.7 5.6 6.6 with Noise Reduction 0.8 1 1.3 1.5 1.8 2.5 Noise generally increases with intensity Adaptive filtering reduces noise by ca. factor 3
Radiometric Quality Image Artifacts N Visible bands in epipolar images
Radiometric Quality Image Artifacts Left Stereo Right Stereo Staircase effect in left image Nonexisting white dotted lines
Radiometric Quality Spilling Strong reflection/saturation Spilling (blooming) Edge sharpening artifacts (overshoot, ringing) Spilling increased due to TDI use Spilling in images over Geneva. Left and middle Ikonos, right Quickbird. The smaller the GSD, the larger the problems. The spill is always in the scan direction (forward in left image, reverse for the other two images). More and larger spills observed with Quickbird than Ikonos.
Radiometric Quality Cause of Spilling Bidec angle (Space Imaging, Eye on Quality, How collection geometry affects specular reflections, 2002)
Radiometric Quality Image Artifacts Left: grey level jumps between CCD subimages ; Right: bright horizontal and vertical stripes
Radiometric Quality Image Artifacts Pan-Sharpened Ikonos Ghosting of moving object due to the 0.5 s time difference between acquisition of PAN and MSI
Radiometric problems - MOMS-02/D2 mission (Space Shuttle) The nadir channel of MOMS consisting of 2 partial CCDs. Their radiometric differences (left) were very large and unusual. Right, shows the result of preprocessing (radiometric equalisation, using the grey values of the 2 neighbouring columns in the middle and assuming that they should show more or less identical grey values; no overlap between the 2 CCDs existed in the delivered image to do a better radiometric equalisation using overlapping regions.
Radiometric problems - MOMS-02/D2 mission (Space Shuttle) Left, original image with missing lines probably due to data transmission problems. These lines could be replaced by averaging of the top and bottom neighbouring pixels. Middle, contrast enhanced that also shows chess pattern noise. Right, the result after preprocessing and contrast enhancement.
CCD Sensors MS channel co-registration and pansharpening CCD sensor: LEFT, one of the spectral images ; RIGHT, pansharpened image (spatial information taken from PAN image, colour information from multispectral images); Example from aerial images but problems can be the same with satellite images, where the difference in geometric resolution between PAN and MS of the same sensor is factor 2-4.
CCD Sensors MS channel co-registration and pansharpening CCD sensor: poor radiometric and geometric quality of pansharpened image at edges (smoothed, color shifts)
CCD Sensors MS channel co-registration and pansharpening CCD sensor: artifacts of pansharpened image ; arrows show color objects that do not really exist
Radiometric Problems Examples from ALOS PRISM, 3-line CCDs, each line consisting of connected smaller partial CCDs 1. Destruction of image signal due to image compression (JPEG-like, compression factor 4.5 or 9, using compression in 8x8 pixel areas and creating artifacts that differ a) between the 3 images and b) even within regions that are homogeneous, e.g. water surfaces) 2. Multiplexing. The even and odd columns of each CCD are first compressed separately, and then the compressed even and odd columns are combined in one image. In addition, neighbouring even and odd columns have other response, and this creates alternating vertical lines. 3. Saturation, espec. at buildings. 4. Horizontal bright lines, possibly due to electronic errors 5. Usual problems like a) small radiometric differences between neighbouring CCDs, b) wrong radiometric calibration of the sensor elements, leading to different response between neighbouring pixels and the creation of vertical stripes Problems 1 and 2 very serious, 3 and 4 less. Making automatic processing of the images, espec. DSM generation very difficult
Radiometric Problems ALOS Prism Different compression artifacts in the 3 Prism channels
Radiometric Problems ALOS Prism Different compression artifacts within one channel, espec. on the right in the homogeneous Thun lake surface. Due to the multiplexing, the compression blocks appear 16 (H) x 8 (V) pixels.
Radiometric Problems ALOS Prism Severe compression artifacts occur often in fields with some regular texture.
Radiometric Problems ALOS Prism Alternating dark and bright columns due to multiplexing.
Radiometric Problems ALOS Prism Channels Nadir, Forward, Backward. Horizontal bright line and rhombus-form bright object in Nadir. Due to electronic problems. There is no bright object in this region as forward and backward channels show.
Radiometric Problems ALOS Prism Again horizontal bright line. Middle: small radiometric jump between 2 partial CCDs. Right: again alternating dark and bright lines due to multiplexing.
Preprocessing - Pre refers to operations BEFORE the actual main processing (e.g. classification) - Aim: improve images to make subsequent processing - easier -faster -Improvement in: - more automated - more accurate and reliable - geometry - radiometry (incl. color) - often two above are related (one can cause the other) - Improvement necessary sometimes due to errors, but not always! - Possible error sources: wrong design (sensor, satellite), electronic noise, compression, crosstalk, multiplexing, use of multiple partial CCDs, jitter of satellite and pointing instability, use of TDI resulting in smoothing, wrong radiometric calibration of sensor elements, firmware (on sat), processing methodology, software (at various stages firmware, user), etc. LONG LIST.
Geometric preprocessing -Various corrections of raw images, e.g. : - correction due to earth rotation - panoramic distortion - resampling - Usually easier to perform, mostly mathematically formulated - Few are difficult, e.g. changes in the interior geometry of the sensor due to temporal differences of the environmental conditions, e.g. shifts, rotations etc. in the focal plane between partial linear CCDs.
Geometric preprocessing - An example of a difficult geometric correction, showing 2 vertical stripes between a reference Lidar DSM and a DSM from matching of Ikonos images, due to a shift of the middle partial CCD of the Ikonos PAN sensor. - This would not be feasible without accurate ground reference data abd accurate image measurement techniques.
DSM generation results (Ikonos triplet, Thun) Height jump of 1.3-1.5 m corresponds to 0.7-0.8 pixel y-parallax error
Ikonos focal plane (shift of middle partial PAN CCD caused jump)
Ikonos interior orientation error The cause of vertical stripes with larger height error due to inaccurate interior orientation modelling. E.g. a possible shift of the middle CCD relative to the other two will cause the same pixel coordinate error for point P1, but not for point P2, introducing thus a y-parallax (and height) error.
DSM generation results (Ikonos triplet, Thun, Dec_N)
DSM generation results (Ikonos triplet, Thun)
- More difficult than geometric one Radiometric preprocessing - Need to know precisely how the image was created (both regarding hardware and software) and think of possible error sources - Need to do the preprocessing (corrections) in the inverse direction as the image was created - Do preprocessing with full number of bits, then reduce to 8-bit if needed - Commercial software often insufficient for such corrections -Same function sometimes means different things in various commercial packages and performs differently (e.g. despeckle in ImageJ is a median filter! And performs very differently than despeckle in Photoshop). Need to study the software documentation but sometimes this does not exist or is very vague (e.g. despeckle in Photoshop). - Two basic classes of operations: restoration (more objective and more difficult), enhancement (easier and more subjective) - To improve defect information, one uses info from neighbouring pixels, and/or info from other channels, preferably taken simultaneously, from the same viewing angle, with the same geometric resolution, and the same spectral properties (practically impossible to do, so always compromises).
Radiometric Preprocessing Aim: Noise reduction, contrast & edge enhancement Methods: 1. - linear reduction from 11 to 8-bit - Gaussian filtering - Wallis filter 2. Like 1 but after Gaussian filtering - unbiased anisotropic diffussion 3. - adaptive noise reduction (2 methods) - Wallis filtering - reduction to 8-bit (histogram equalisation or normalisation)
Original Preprocessing - Noise reduction, contrast & edge enhancement Original, contrast enhanced Preprocessed 2 Preprocessed 3
Edge preserving noise reduction with adaptive fuzzy filtering (right). Small details are kept and edges are in addition sharpened (Pateraki, 2005).
Leibniz An example of enhancement (Ikonos, Thun) Contrast Enhancement Shadow Area Original IKONOS Image Enhanced IKONOS Image with Wallis filter XXIst ISPRS Congress, Beijing, China, Tutorial-10, July 3rd, 2008
Reduction to 8-bit. Left with linear transform, middle histogram equalization, right histogram normalization (Pateraki, 2005).