GISdevelopment.net ---> AARS ---> ACRS 1992 ---> Digital Image Processing

An algorithm of extracting contours to produce DTM from muti-color topographic map

Guo Jun,Zhu Chongguang
Institute of Remote Sensing Application Chinese Academy of Sciences
P.O. Box 775, Bejiing 100101, P.R. China


Abstract:
Multi-color topographic map is an important information resource of Geographic Information System. Digitizing contour lines is one method of obtaining DTM. We can obtain color map R,G, and B from color scanner. Extracting one color from color spatial cube is actually a spatial clustering. Because it is not satisfying of the quality of topographic Map and distribution of color, the result of common method of classifier may not be satisfied. Appropriate spatial transformation can be used to obtain better result. In this paper, based on scanning digitizing, we present a serial method of extracting contours from multi-color topographic map, and followed processing to produce DTM.

Introduction:
At present, single element is extracted from multi-color topographic map by means of manual tracking digitizing whose efficiency is very low especially in case of complex, large amount data and long processing time. to overcome the disadvantages, we digitize topographic map with scanner. There are two ways to digitize contour lines with scanner. The first is monochromical scanning through which 8-bit images can be obtained. The second is color scanning. We can obtain 24 bits image of Red, Green and Blue.

For digitizing topographic map to produce DTM, it is necessary to extract contour lines form multi-color topographic map according to different colors. Extracting contour lines means that one color is separated from color space of RGB. Then we remain necessary color and remove unnecessary colors. The work is essentially a spatial classification. The method of minimum of distance and maximum likelihood are usually used. In fact, because the color of map Red, Green and Blue is not true to the original color and the distribution of color may be cross, common method couldn't obtain better result.

In this paper, we present a method of spatial transformation to improve the distribution of the area of color so that some color can be easily separated from other colors. There are several steps as following.
  • spatial transformation.
  • Stretching transformation.
  • Slicing one color from the space of transformation.
The map which only includes contour lines will be processed by several method of Mathematical Morphology [1], such as removing noise, thinning, assigning, interpellation, etc.

The description of the algorithm of extracting one color form RGB space:
  1. The Generation of Color Spatial Model
    After color scanning, we obtain three bands of Red, Green and Blue. Because the three bands are separated, we could regard them as three perpendicular components. Let's generate a color spatial cube shown in Fig.1.


    Fig. 1 Color Spatial Cube

    The original point of color cube corresponds to Black (whose value of r,g, and bare equal to zero). The eight vertexes of color cube correspond to eight full color area respectively.

    When color cube is generated, every point's color of topographic map corresponds to identified vector in vector space. The identified vector corresponds to an identified color.

    Because of the difference between background colors, hue and precision of scanner, every color in multi-color image has inhomogeneous distribution. There are some difference between different point with the same color in the hue, intensity and saturation. Every color in vector space corresponds to a vector group. The more homogeneous the color is, the smaller the color area is. Inversely, the more inhomogeneous the color is, the larger the area of color group is. The area of color are sometimes cross with each other, and sometimes not.

    If, of the three bands of R, G, B in base color A, B, there is at least one band not to cross with others, A, B, the two color fields are not connected. Only if the three bands of A, B cross with each other, A and B are connected.

    Essentially, algorithm of extracting contour lines from R, G, B images is to separate one color form others. As we know from Topology, whatever transformation is to be sued, if area A is separated form others, then color A should be extracted.

  2. Spatial Transformation
    Now we present an algorithm of spatial transformation which can transform R,G,B to another space. Because the areas of RBG cross near to each need to transforming. The aim of transform is that distance between the necessary colors and unnecessary colors should be put away.

    The formula of spatial transformation is:



    Every color has its own area of saturation. Since the extent that the color fields are stretched to their saturation is different, variable Landsat is used to control, the stretch extent. Ak, bk, ck, dk are related to the stretching direction. They can be suitable chosen to separate determined color field A from other fields in color cube.

  3. Stretching
    To make the grey levels in some range compressed, or stretched, corresponding non-linear transform can be done on every and after spatial transformation. There is logarithmic transform:



    The topological space has three axes W1, W2, W3 after the transform above. The point in color cube are stretched or compressed but the connectivities are kept same.

  4. Slicing
    To separate "necessary color field" completely without unnecessary color, the best method is to use some spatial surface to slice color cube. Simply, inclined planes consisting of the linear combination of three channels can be used.



    To separate different color field satisfyingly. The simplest and most convenient method is to use the three planes paralleled to the axes:





    There may be some noise in the result, such as interrupted points which are overlayed by characters or kilogram grids. So, after the process described above, some work has to be done to remove the noise, and connect the interrupted points, etc. Mathematical Morphology is one of the good methods solving those problems.

    The noise of original image RGB might be trouble for processing. Appropriate preprocessing is necessary. The distribution of noise is random. Using principal component analysis for dividing original image into principal component and noise component. After KL transformation has been used, three principal component were obtained. The first principal component includes most information of original image. the information included in the second and the third principal component are then less and less.

    Because KL transformation has no effect on unrelative noise, the last component include most of noise of original image. Appropriate processing can compress noise.

    After compressing noise and ratio processing with different bands, the result can be used as reference image in the procedure of slicing.

    After processed with the ways mentioned above, the image may have some isolated points and unnecessary short lines. Then use morphology's dilation and erosion and other combining operations can be used to remove noise and obtain better smooth binary image of contours.
Thinning, Modification and Interpolation
  • Processing of contours. The noise are usually isolated point. The map of binary image will be filtered to remove noise.

    Sequential thinning in morphology will be used. Let A be an binary image, S (A) the result of thinning.

    S (A) = (A. {Bi} ) m (6)

    Here, i = 1,2,..,8.

    .is thinning operation symbol of morphology,
    { } is sequential operation symbol of morphology.

    Bi is structural elements.


    m is the maximum number of iteration.

  • Discontinuous point processing.
    Using hitting operation in Morphology to find the discontinuous point. We can trace every contours to find discontinuous point. Adding heuristic information, we can use the direction code as prior direction in the deep first search. After finding corresponding points, we can judge its continuity so as to connect the two points.

    If the result is not satisfying, we can add manual operation. After the contours is assigned, we interpolate it to produce DTM.
Producing 3D Model from DTM.
In order to display 3D image, we need to get the date of image which has been registered with DTM. The topographic map of R.G.B. should be registrated with the TM image. When we select point pairs from the topographic map and TM images, the selected point should have invariable characteristic, and have homogeneous distribution. Otherwise, the result of geometric registration are deformed on the edge of the image. The model of geometric correction is:


The essence of the procedure from 3D stereo model of DTM of 2D displaying is perspective transformation.

S (sx, ys, zs) is set to be view point. The object point (x,y) can be counted from following the formula:


After obtain two dimensional coordinates, the image should be processed by hidden operation and integrated with TM image. Finally we obtain three dimensional display of spectrum image.

Experiment and Conclusion
Processing of multi-color topographic map by using spatial transformation can reduce the work time of manual digitizing. The color information of multi-color topographic map is much more than that of monochromical map. Spatial transformation is different from spatial cluster. Spatial cluster is that the points which have determined distribution can be recognized and distinguished. The spatial transformation used in this paper is tried to change the distribution of spatial points. Not only the points around the area of color are contracted, but also the area of color is moved. For these reasons, results of spatial transformation is better than spatial classification. Certainly, because the precision of scanning topographic map is limited in high precision, the request of precision is at clears 500 dpi. Good result can be obtained under appreciate conditions, such as high precision scanner and topographic map with better quality. We chosen a topographic map for the experiment in c4500 scanner. It's higher precision is 25u. By the processing mentioned above result can obtained.

Reference:
  • J. serra: Image Analysis and Mathematical Morphology, Academic Press, New York, 1982.
  • A. Rosenfeld, A.C Kak, Digital Picture, Processing, Academic Pres, 1976.
  • Theo Pavlidis: Algorithms For Graphics And Image Processing, Computer Science, Press, Inc. 1982.
  • David F. Rogers: Procedural Element for Computer Graphics, McGraw - Hill, Inc., 1985.