# Remote Sensing Image Rectification and Restoration Image Rectification

• Slides: 29

Remote Sensing Image Rectification and Restoration

Image Rectification and Restoration ► ► ► Geometric correction Radiometric correction Geometric restoration

1. Geometric Correction For raw image rectification ► For multi-date images registration ► For multi-resolution images or data layers registration ► ► Systematic distortion vs. random distortion

Skew Correction Coordinate transfer Pixel value resampling http: //rst. gsfc. nasa. gov/Intro/Part 2_15. html

Ground Control Points (GCP) Features with known locations on a map (X, Y coordinates). These are the “ground control points” ► The same features can be accurately located on the images as well (column, row numbers) ► The features must be well distributed on the map and the image ► Highway intersections are commonly used ground control points ►

Finding UTM coordinates on a map

Coordinate Transform ► Coordinate transform equations relate geometrically correct map coordinates to the distorted image coordinates x = a 0 + a 1 X + a 2 Y y = b 0 + b 1 X + b 2 Y x, y: column, row number X, Y: coordinates ► Root Mean Square Error (RMSE) = √(dx)2 + (dy)2 Calculate RMSE for all control points

Resampling ► The purpose is to assign pixel values to the empty pixels in the rectified matrix output ► Superimpose the rectified output matrix to the distorted image ► The digital number (DN) of a pixel in the output matrix is assigned based on the DN of its surrounding pixels in the distorted image

Re-sampling Methods ► Nearest neighbor resampling ► Bilinear interpolation ► Cubic convolution resampling

Nearest Neighbor Resampling ► The DN of a pixel in the output matrix is assigned as the DN of the closest pixel in the distorted image ► Advantages simple computation maintain the original values ► Disadvantage spatial offset up to 1/2 pixel

Bi-linear Interpolation Distance-weighted average of DN values of the closest 4 pixels ► Advantage output image is smoother than the nearest neighbor method ► Disadvantage alters the original DN values ►

Cubic Convolution Resampling Uses DN values of the closest 16 pixels, adjusted by distance ► Advantage smooth output image ► Disadvantage alters the original DN values ►

When to Rectify before image classification ► Rectify after image classification ►

2. Radiometric Corrections Radiometric responses differ by ► dates ► sensor types ► images ► Causes: - Illumination - Atmospheric conditions - View angle or geometry - Instrument response

Radiometric Corrections ► Sun elevation correction ► Atmospheric correction ► Conversion to absolute radiance

Sun Elevation Correction DN ► ----------------Sin (Sun elevation angle) ► Assuming the terrain is flat

Satellite Summer Zenith Solar Elevation Angles Spring / Fall Winter Tangent plane Equator

Atmospheric Correction Haze compensation The DN value of an object (e. g. , a deep clear water body) with 0 reflectance = Lp ► Subtract the DN from the entire band ►

Absolute Irradiance Conversion of DN values to absolute radiance values ► It is necessary when compare different sensors, or relate ground measurements to image data ► L = (Lmax- Lmin)/255 * DN + Lmin ►

3. Geometric Restoration Stripping ► Line-drop ► Bit errors ►

Striping ► Malfunction of a detector ► Use gray scale adjustment to correct the strips

Line Drop ► using average of the above and below lines to fill the dropped line

Bit Error Salt and pepper effect due to random error ► Use 3 x 3 or 5 x 5 moving window average to remove the noise ►