Production of 1:50,000 scale image maps using satellite imagery

Production of 1:50,000 scale image maps using satellite imagery

Kasi Balarathinam
Digital mapping centre, Survey of India
17-EC Road, Dehra Dun-248001, INDIA

Abstract
The author has the privilege of creating an image map, by combining SPOT panchromatic data with the Landsat TM data and annotating with selected topographical map features. Bands 5 3 1 of the TM data are used in the R G B (Red, Green, Blue) channels in false colours. The work is done in the International Imaging System(IIS) using the System-600 software in VaxVms environment . The job is scheduled to be completed by Aug. 90 and the map printed by sept 90. Brief-600 command history as used in the process with description of the purpose, is appended as enclosure-1.

The TM image is registered with the reference SPOT Panchromatric image using 28 ground control points (GCPs). By using least squares regression technique(4), the RMS error in the image registeration, could be reduced to less than 0.9 of a TM pixel. The SPOT and the TM data are mixed by adding the intensity values of the SPOT pixel at each location to the intensity value of the TM pixel in the corresponding location in each of the three bands. The output image is scaled suitably and registered with the 1:50,000 topographical map using 21 GCPs achieving an RMS error of less than 0.9 of a SPOT pixel (9 m ) . The information contents in the SPOT panchromatic data of 10m resolution are found enhanced in the multispectral environment in the TM bands reasonable success.

The image map is annotated with selected topographical map features. Contours at 200m vertical interval, annotations and other cultural information (not clear in the image), are compiled from the topographical maps and composed with the image map in suitable R G B colour values. Contours are merged with the image map in blue and brown for regions of permanent snow and other regions. The permanent snow area is classified (using the minimum distance classifier) and filtered (using mode filters of varying kernels in succession). The resulting snow region in raster form is edited in the Intergraph environment. This classified and processed output is used as a mask to split the contour file of the topographical map into two files viz, one for the permanent snow bound region and the other for the remaining regions, using map composing techniques.

In principle, the primary colours in the R G B channels in the video, are identical to the negated C M Y (Cyan, Magenta & Yellow) reprographic inks. The optronics 4040 photo plotter can directly plot or negate and plot these R G B band images with suitable half-tone screens for reproduction in C M Y.

This beginning in the image map reproduction in the survey of India, is a mile stone in the mapping technology for planners, engineers, scientists and other users in India.

The input satellite data for mapping
The SPOT panchromatic data and Tm data for the region of mapping were supplied by the national Remote sensing Agency (NRSA) . The area for mapping is incident in adjoining TM quadrants of a scene in the X-direction and the IMAGER, an Intergraph image processing system, has the necessary software to combine the adjoining quadrants situated in the X or Y direction. We, therefore, read the TM data in the IMAGER. By examining the TM image, Bands 5 3 1 were selected for projection in false colours in the R G B channels.

Registeration of TM image with SPOT image
The SPOT and the TM images were displayed in a split screen mode and 33 common points (GCPs) were chosen. A polynomial least squares fit the RMS error as 2.015 TM pixels and indicated that the 17th GCP is the most erroneous point with error as 6.889 pixels. This erroneous control point was deleted. Now the next fit agreed with lesser RMS error. This erroneous control point was deleted. Now the next fit agreed with lesser RMS error, This process of least squares regression was repeated until about 12 control points were left. The analysis indicated that 18 control points would be and optimum choice, with RMS error as 0.896 TM pixels. The Warping is done using full bivariate third order Legendre Polynomial. The rate of convergence of the RMS error will be very slow if the recorded machine coordinates are rounded and not precise. It is avoided by enlarging the region of incidence of the control points and indication the location as precisely as possible. The rate of convergence is then found to be faster.

Merging SPOT Panchromatic and Landsat TM data
The Landsat TM image is now a layered image over the SPOT panchromatic image. Digital numbers (DN) associated with the Landsat TM image may be combined or merged with those for the panchromatic reference image using techniques as previously referred by R Welch and Manfred Ehlers[1] (Photogrammetric Engineer and Remote Sensing, vol53, No3 March 1987, pp 301-303) over the work of saint and well Cliche et al., and Chave (1984-86). These methods may be summarized in the following equations:

Mi = Ai * SQRT ( M'i *P ) + Bi .........................(i)
Mi = Ai * ( W1 * M'i + W2 * P ) + Bi ........................(ii)

Where Mi is the DN for a pixel in the i-th band of the merged image; M'1 is the DN for the corresponding pixel of the i-th multispectral image; P is the DN for the corresponding panchromatic reference image pixel; W1 &W2 are weighting factors and Ai and Bi are scaling factors to place the resulting DN within the dynamic range (0,255) . The author used the second algorithm with responsable success. Unit weighing factors and default scaling factors as in the software for pieces of 512x512 size, were used. Performing this job for each 512x512 frame in piece meal for the full image, is best done by designing a system-600 command procedure.

Registering the merged image with map
Polyconic projection over the modified Everest spheroid is used for the image map. The central latitude and longitude coordinate pair (in degrees), for the region of mapping is assigned the equivalent Lambert Grid Coordinate pair. The software can now output and internal grid coordinate pair for each input coordinate pair in degrees of latitudes and longitudes for the sheet. The internal grid coordinate for the limiting graticule corners are obtained in this way for subsequent identification during georeferencing.

The existing 1:50,000 scale topographical maps were used for selection of Ground Control points (GCP). The map is mounted in the Intergraph environment for digitisation of GCPs. The corner coordinates in degreed for the latitude and longitude of the map are entered to obtain the coorinates of the GCPs in same terms as needed in the IIS environment. The GCPs are selected from the image and digitized over the map to obtain their degree coordinates, since our IIS has no digitizer in the configuration. These coordinate are fed into the IIS by keying in over the terminal.

Same technique of least squares regression as described before is used to attain optimum selection of GCPs. We used 21 GCPs with RMS error as 0.853 SPOT pixels . The warping is done using full bivariate fourth order Legendre Polynomial . The image is then warped to register with map coordinate system. For both registrations the error status as found is appended in the enclosure-2.

Geocoding the warped image
Internal grid coordinate values for the corners of the sheet of mapping is obtained from the machine after we input the Lambert Grid values for the central latitude and longitude of the sheet as described in the previous para. These values are easily located in the warped image. Now the monitor will show the column line coordinates of the corners of the sheet for mapping The georeferencing graticule tick marks are placed in these locations.

Superimposition of map data over the warped image
The following topographical map layers processed in the Intergraph environment are used for superimposing over the image map.

Contours at 200m vertical interval
Settlements and lines of communication
Glacial features.
Annotations.

The layers are required to the rotated to make the parallels of latitude parallel to the X-axis of the machine. The IIS software on the other hand rotates the image at the warping stage achieving the same. Therefore the internal corner grid values for the sheet of mapping as given by the machine and as may be obtained from the Lambert Grid Tables, will differ and it will have no effect in the placement of latitude and longitude ticks.

Splitting of contour file in IIS environment
The glaciated and permanent snow regions are depicted with contours in blue and other regions with contours in brown. This necessitates splitting the contour line file into two separates. It is done by supervised classification of snow bound area and using the solid snow area as a mask to clip out the contours incident over the bound area, from the rest of the file.

A minimum distance classifier is used for the supervised classification. The output from the classifier is used as input for mode filtering with as 5x5 kernel of unit weight. The filtering process is repeated with a 7x7 and 9x9 kernels recursively. If still the fragments of larger size, but not larger enough to snow as snow bound areas persist, raster edit facility in the Intergraph environment is used to interactively clear them as noise. Fragmented snow mask and holes in snow mask are unpainted and painted as snow respectively during the raster editing. The IIS raster edit facility, for this purpose in the existing modules, is cumbersome to use. The output from this process can extracted as a filled area in a single band image as a positive snow mask.

The contour file of the sheet of mapping is a line file in single band. We convert it as a multispectral image by merging its 3 copies in R G B bands, if we use this multispectral output as an optional back ground for composing the negative of snow mask with colour values as (0,0,0) for the (R G B) channels the contours under the snow mask in the multispectral image alone will remain and the balance became dark in the output image alone will remain and the balance became dark in the output image as per colour value (0,0,0). Any one of the three bands from the resulting output will serve as a line file for contours in snow bound area. By repeating the above process for the positive of snow mask we can get the complement of the contour file representing the area not snow bound.

Map Composing
The following line files from the topographical maps as complied in the Intergraph environment are composed into the (SPOT +TM) image as per colour values indicated therein. The intergraph design file in vector format is converted to the IIS image file in the raster format using the IIS link software. By suitably selecting the (column, line) size for the output raster file between the maximum and minimum coordinate values in the input vector file, the rasterisation of the vector file to any pixel size in the raster format can be achieved. In this case we achieved a 10 m pixel for printing at 5 pixels per mm on scale 1:50,000.

Sl.	Description of file	Colour (R,G,B) %
1.	Contours in snow bound region	20	20	80
2.	Contours in other regions	56	42	14
3.	Glacial features	0	0	0
4.	Settlement & communication	100	0	0
5.	Annotation of text	0	0	0

Reprography [3]
The primary colours R G B can be represented as below for reproduction in printing colours Cyan, magenta and yellow :-

Cyan = G + B = 1- R
Magenta = R + B = 1 - G
Yellow = G + R = 1 - B

The effect of cyan is possible to be achieved by negating the red band and printing the output in cyan ink. Similarly we negate green and blue bands and print in magenta and yellow inks green and blue bands and print in magenta and yellow inks respectively. For example if an area in an image contains only one red information in a black surrounding, then the process printing stages can be represented as below :-

The work area is excluded for cyan printing and included for the M+Y printing which will give an output in red. The 'no work' area will receive C+M+Y impressions to give an output in black. For sharpening the black information such as the NAME, we can have a separate black plate. This is the status needed for the reprographics.

In this case we created the following files for plotting in the optronics 4040 plotter for reprography. The Optronics can interpose desired half-tone screens and plot the file as positive or as negative, for preparing of contact plates for reproduction in C M Y B (B-black) plates.

Red band image as positive
Green band image as positive.
Blue band image as positive
Annotational texts as negative.

Conclusion
With improvement in image processing techniques, such as transformation RGB vs IHS[2}, RGB vs YIQ , the lineament in the SPOT data will stand out with enhanced interpretability Creation of permanent GCP vector files for all types of imagery on 1:250,000 scale sheet basis, using photogrammetric coordinates will go a long way in a production environment for quality registration . In such case the image map superimposed with topographic map data can serve as a map for all users. The output achieved is based on the techniques experimented and on the materials is based on the techniques experimented and on the materials immediately available. Over the course of time the quality of image maps will improve and the image maps may replace the conventional line maps. The advantage in image mapping is that no drafting is involved and the production of image maps can be faster and up-to-date. Image interpretation skill will enhance the utility of the image map.

Acknowledgement
The author is very grateful to Lt Gen S M Chadha and Mr. S D Baveja for having entrusted this prestegious opportunity of producing the first ever image map in the survey of India to the author. The unique co-operation I received from Mr. Meena, George and Chinnawar is something, I have never had before, and I am indeed grateful. This high-tech image mapping became a reality because of the valuable guidance by Lt Col KK Naithani, Dr P Stefanovic and Mr. Kostwinder and the author feels indebted to their contribution.

Reference

R wilch and Manfred Ehlers, Laboratory for Remote Sensing and mapping Science, Department of Geography, University of Georgia GA 30602
Mohammed Essadiki in ITC Journal 1987-1, A combination of panchromatic and multispectral SPOT image for topographic mapping, Institute Agronomique at Veterinaire Hassan, BP 6202, Rabat, Marocco.
Kasi Balarathinam, Defence Electronics Application Laboratory seminar on 12 Apr 90, of title same as above , from the modern Cartographic centre, Dehra Dun- 248001
Kasi Ba;arathinam and Sachidanand Semwal , INCA seminar of Nov 89, Geometric correction of remote sensor data using least-squares regression procadure in IIS, Modern Cartographic Centre Dehra Dun- 248001.

Error Status in image and map registration: Enclosure - 2
Image registeration				Map registeration
GCP	X-error	Y-error	Hypot	X-error	Y-error	Hypot

1	0.238	-0.367	0.437	-0.143	-0.956	0.967
2	-1.034	0.357	1.094	-0.174	-0.156	0.234
3	0.588	0.013	0.588	0.137	0.011	0.137
4	0.218	1.659	1.673	0.000	-0.011	0.412
5	0.537	-1.230	1.342	0.328	-0.130	0.353
6	0.197	-0.276	0.339	-0.826	1.280	1.523
7	-0.641	0.371	0.741	-0.140	1.078	1.087
8	-0.005	-0.461	0.461	0.654	-2.102	2.201
9	-0.155	0.882	0.895	0.387	0.063	0.392
10	-0.159	-0.223	0.274	-0.478	0.187	0.513
11	0.603	0.287	0.668	0.497	0.098	0.506
12	-0.556	0.142	0.574	-0.649	-0.262	0.700
13	0.688	-1.387	1.548	-0.105	-0.210	0.235
14	-0.166	0.195	0.256	-0.075	0.204	0.217
15	0.176	-0.193	0.261	0.783	0.251	0.822
16	-0.598	-0.007	0.598	0.476	-0.106	0.488
17	-0.873	-0.302	0.924	-0.348	0.358	0.499
18	0.098	-0.116	0.152	-0.634	-0.028	0.634
19	-0.383	0.201	0.433	-0.160	0.145	0.216
20	-0.404	0.469	0.619	0.028	0.744	0.744
21	0.632	-0.621	0.887	0.452	1.417	1.487
22	1.522	0.384	1.570
23	1.026	-0.012	1.026	Image registeration by 3rd order fit
24	-0.234	-0.578	0.623
25	-0.345	0.734	0.811
26	0.770	-0.807	1.115	Map registration by 4th order fit
27	0.177	0.638	0.662
28	-1477	0.250	1.498

Max Errors			1.673			2.201
Signed Mean			0.788			0.684
Abs Mean			0.788			0.684
RMS Error			0.896			0.853