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

Feature extraction from remotely sensed data using modified Homomorphic Filtering Approach

Nitin Kumar Tripathi,K V G K Gokhale
Department of Civil Engineering
Indian Institute of Technology
Kanpur - 208016, India


Abstract
It is often seen that due to spectral and spatial constraints as also neighbouring environmental effects, some features or details on Remote Sensing imageries are not so prominent that they can be mapped effectively. Several image processing techniques have been used by various researchers based on features of interest. A digital image processing technique discussed in the present study is very handy as it brings out the various hydro-geomorphological, lithological and landuse / landcover features which otherwise are not distinct on Remote Sensing imagery. As a case study, the processing of the IRS-1B data of Varanasi (India) area in the present work has been undertaken. Use of this approach has enabled delineation of confluence of rivers Ganga and Varuna, not distinct on imagery. In addition, boundaries of sand bars, oxbow lakes, meander patterns, flood-plains, paleochannels, faults, soil types and minor land use / land cover features are demarcated clearly. Some of these features do not appear distinctly even after processing using traditional image processing techniques. The digital images obtained using present methodology are of immense value is preparing a detailed and accurate map of the area for varied applications.

Introduction
In recent years, digital image data have been considerable use in various applications. Digital image, processing techniques generally used area image enhancement, edge detection, feature extraction, segmentation, image coding etc. In the present paper an attempt has been made to evolve a digital image processing approach which can improve the contrast of the satellite imageries to the extent that most of the lithological, hydromorphological and landuse / landcover features which are not so distinct become clearly visible on the output imagery.

In the process of evolution of present methodology various other digital image processing techniques like Robert's Operator, Sobel Operator, high results of these techniques are compared with the current modified homomorphic approach. Homomorphic is a term adopted from the Abstract Algebra to describe a transformation between algebra groups that preserve linear combination (Dubisch4). In general an image of large dynamic range i.e. a natural scene of earth surface on bright sunny day is recorded on a medium with small dynamic range such as film or photographic paper. This is the cause of significant reduction in mage contrast mainly in brighter and darker regions (Lim1). In order to enhance the image one approach is to reduce its dynamic range and enhance its local contrast before recording it on a medium with small and limited dynamic range (Gonzalez6). The other approach which is followed in present work is to diminish the dynamic range to certain extent, preserve the contrast and introduce a nonlinear component of dynamic range in final output. Several others have earlier used homomorphic approach on Landsat data and according to them "Homomorphic Transforms that permit realistic linear enhancement of Landsat images can lead to a superior product for final interpretation" (Carrol2).

The Study Area
The area selected to test the present image processing algorithm is Varanasi, India. The path and row number in IRS (Indian Remote Sensing Satellite) coverage is 23 - 50. The data is acquired in digital form on a CCT in April 1988. This area offers a scope to test the present methodology in a all respects as it consists of various type of hydrogeomorphological and landuse / landcover features. Ganga and Varuna are the two rivers flowing through the Varanasi urban area.

Methodology
From time to time homomorphology technique has been used by many signal processing and image processing workers for various purposes. Homomorphic filtering (Oppenheim3) is a useful technique for image enhancement when an image is subjected to multiplicative or interference. This technique can be very useful for multispectral Remote Sensing data specially in a scene where vital information contents are lost due to severe cloud effect, radiometric constraints, hue effect etc. These factors influence the dynamic range to such an extent that important details are lost or become indistinct. As already briefed earlier homomorphic approach can be one such approach to accomplish this task. One simple image model is as follows :

f (n1, n2) - i (n1, n2) r (n1, n2)

where
f (n1, n2) represents image
i (n1, n2) represents illumination component
r (n1, n2) represents reflectance component


In the process of developing a homomorphic system for Image enhancement, the illumination considered as the primary contributor to dynamic range of an image varies slowly. The reflectance component is primarily contributor to the contrast and responsible for distinguishing among numerous details and patterns. It has the tendency to vary at a faster rate.

To modify these component the first task is to separate them. One approach is to take the logarithmic of above image model and obtain:

log f(n1, n2) - log i (n1, n2) + log r (n1, n2)

The pass filtering will separate the dynamic range while high pass filtering will separate the contrast component. Now the task of attenuation of dynamic range and enhancement of contrast level can be taken up. Exponentiation after filtering completes the enhancement process.

In present scene of IRS 1B LISS 1 image, the confluence of the rivers Varuna and Ganga is not at all distinct in any spectral band owing to reason that neighbouring dense urban landuse features, other details and atmospheric effects have obscured the confluence point. This kind of interference is quite common when the major dense details suppress the minor details. This kind of intermixing and interference is to be eliminated in order to restore the important minor details present in the scene.

In the approach evolved to solve this kind of a problem, the image f (n1, n2) is smoothened. The smoothened image which is the dynamic range component in this case is subtracted from the original image to obtain the high frequency component responsible for contrast and identification of various features. In the smoothened image, a nonlinearity in the form of exponential function is introduced. This component is added to the high frequency to get the final image.


Figure 1 The proposed homomorphic image processing scheme


Figure 2 Original IRS 1B LISS1 Scene of varanasiand surrounding area


Fig 3 Image After Neighbourhood Averaging


Fig 4 High Frequency contents of the original image


Fig 5 Output image after processing by modified homomorpic filtering approach

Results and Discussion
The IRS 1B LISS 1 scene of the study area is shown in figure 2. It shows course of Ganga with hydrogeomorphological features such as meandering patterns, oxbow lakes, sand bars, flood plains. The urban area of Varanasi is visible on the left bank of river Ganga and small urban fringe of Ramnagar on the right bank of Ganga. Some road network and railway lines are also seen. The right portion of the image is comparatively too dark than the left portion. From topographical map it is clear that an important detail is almost vanishing from the satellite imagery i.e. course of river Varuna and confluence of this river with Ganga. Other salient details pertaining to hydromorphology are also missing.

The original image is smoothened using neighbourhood averaging algorithm and the output image is shown in figure 3. This is the dynamic range of the original scene. When this part is subtracted from original image (fig.2), the high pass filtered image of the original is obtained. Figure 4 displays the contrast or high frequency contents of the image. in the figure 3 a nonlinearity function is introduced. This image is added to the high passed image (Fig. 4) to get the final image as shown in the figure 5.

The final output image is remarkable clear and this distinctly shows the course of river Varuna and its confluence with the major river of the area i.e Ganga. The hydromorphological features such as sand bars, meandering patterns, paleo channels and flood plains exhibit more details compared to the earlier original imagery. Sand bars on the original scene are displaying some vegetation features, inland water etc. The smaller sand bars are also visible which were almost obscure in original image. The boundaries of all hydromorphological features are also nicely demarcated. Flood plains depict greater details than earlier original image. The urban dwellings also appear clearly. Road networks and railway lines linking Varanasi and Ramnagar cities are clearly visible.

Conclusion
Digital image processing has an immense potential in the application of Remote Sensing for better resource evaluation and information extraction. The technique discussed has presented very encouraging results for pattern recognition and identification of various features which owing to atmospheric effects and intermixing nature of details in the satellite imagery are not visible for visual identification and mapping. The modified approach of homomorphic image processing offers an excellent scope for retrieving important resource information from satellite imagery which is affected by cloud cover, haze, intermixing nature of details and blurring.

References
  • Lim, J.S., "Two Dimensional Signal and Image Processing", Prentice Hall.
  • Carrol, S. and Robinson J. F., "Homomorphic Processing of Landsat data", Canadian Journal of Remote Sensing, 1977, Vol. 3, No. 1, December, pp 66 - 75.
  • Oppenheim, A.V. Schafer, R.W. and Stockham, T.G. (Jr), "Nonlinear filtering of multiplies and convolved signals", Proceedings of IEEE, 1968, Vol. 56, No. 8 pp. 1264-1291.
  • Dubisch, Roy, "Introduction to abstract algebra", 1965, John wiley & sons, New York, 193 p.
  • Gonzalez, R.C. and Wintz, Paul, "Digital Image Processing", 1977, Addison Wesley Publishing Company, Inc.,
  • Pratt. W.K. "Digital Image Processing", 1978, John Wiley & Sons. Inc.
  • Peili, T. and Lim J.S., "Adaptive Filtering for Image Enhancement", Opt. Eng., 1982, Vol. 21 pp. 108-112.
  • Lim, J.S., "Image Enhancement", Digital image processing Techniques", Michael P. Ekstrom (Ed.), Academic Press.