Wavelet based Soil Analysis and Characteristics
SOUNDARARAJAN EZEKIEL
ROBERT A. HOVIS
Ohio Northern University
ADA, Ohio, USA
and
Rick A. Robbins
Ohio Department of Natural Resources
Findlay, Ohio, USA
ABSTRACT
A wavelet based feature extraction method was developed and its application to aerial photographic images on Pleistocene-age glacial deposits in west-central Ohio is investigated. Energy calculation is used to correlate surficial soil properties and characteristics. In our approach, four regions of interest (cultural areas, hydrographic areas, vegetative areas, and bare soil surface areas) are extracted. The method is therefore expected to enhance image analysis of surficial soil images. Multiresolution Analysis (MRA) allows the multiscale analysis of images and thus the extract edges and textures of the images. The wavelet transforms enables the modeling of information at different characteristic scales. We show that the proposed method extracts certain texture features. The method is therefore expected to enhance image analysis of surficial soil characteristic studies.
Keywords: Wavelet, Multiscale, Multiresolution, Extraction, Surficial Soil Image.
1. INTRODUCTION
The current development of wavelets [1] and Multiresolution analysis (MRA) provides a new and powerful tool for digital image processing and pattern recognition. There is already an impressive list of applications, including image processing and video compression, automatic target recognition, image enhancement, edge detection, noise reduction and texture analysis. We apply this for surficial soil properties and characteristics. Although the first works on analyzing satellite images have been initiated in the seventies, few of them has been concerned by urban areas. Indeed, the production of reliable method is difficult because of the complexity of urban landscapes. This complexity depends both the spatial resolution of images and the inner organization of urban areas. Then, the finer the spatial resolution, the easier the discriminations between agricultural fields [2]. The analysis of aerial photographic imagery provides an excellent opportunity to develop the techniques and methodologies to study and correlate surficial soil properties and characteristics of an individual agricultural field. Agriculture is an important industry in Ohio. The success of the agricultural industry is dependent on many variables such as crop type, soil moisture, nutrient availability and soil type. Many of these variables affecting field crops can be monitored using remotely sensed data throughout the growing season. Ultimately, satellite data should be able to provide accurate up-to-date information on agroecosystems at a regional scale. Our paper presents a new method to extract and analyze surficial soil properties which is based on both MRA and wavelet transform. The paper is organized as follows. In Section 2 , we review wavelet based multiresolution analysis. In section 3, we present our methodology for texture feature extraction from images. In Section 4, we illustrate our method and its use by presenting some experimental results. Finally, in Section 5, we summarize and conclude our work.
2. WAVELET BASED MULTIRESOLUTION ANALYSIS
Wavelets are presently used in many disciplines of science and engineering. In this section, we present the concept of wavelets, Multiresolution analysis, and texture segmentation by using MRA and wavelets. The development of wavelets resulted from the need to generate algorithms that would compute compact representations of functions and data sets at an accelerated pace. In the last few years, the wavelet transform has become a cutting edge technology in the image-processing field.
Wavelets and MRA
Jean Morlet and Alex Grossmann introduced the concept of wavelets. It was mainly developed by Y. Meyer [7]. Stephen Mallet developed the first algorithm in 1988[8]. After that, many scientists like Ingrid Daubchies and Ronald Coifmen contributed to this field. A wavelet is a waveform of effectively limited duration that has an average value is zero. So, wavelet analysis is done by breaking up of a signal into shifted and scaled versions of the original (mother) wavelet. From this, we can define a continuous wavelet transform as the sum over all time of the signal multiplied by a scaled and shifted version of the wavelet function . i.e.
where scaling means stretching(or compressing) and position means shifting the wavelet.
The two-dimensional fast wavelet transformation of a image is computed using the same scheme as used in one dimensional case. First, each row of the image array undergoes decomposition, resulting in an image whose horizontal resolution is reduced by a factor of two and whose scale is doubled. The high-pass component of the decomposition characterizes the high frequency information with horizontal orientations. Next, the high-pass and low-pass sub images obtained by the row decomposition are each separately filtered column wise to obtain four subimages corresponding to low-low pass, low-high pass, high-low pass, high-high pass row-column filtering. The later three parts captures the horizontal, vertical and diagonal features, respectively, of the image[10]. After decompositions of the image into low-pass subimage and three high-pass subimages as described as above, the low pass subimage is again subject to row-column filtering operations to obtain a further (coarser) decomposition, and this process is repeated either until the low-pass image has no more interesting features or a desired number of times.
Texture Segmentations
One of the best methods for texture characteristics was, many years ago, multi-channel filtering using Gabor filter[4]. The problem with this technique is not considering all the harmonies of the texture so the performance of the segmentations drops dramatically. Recently many methods are proposed as an alternative for texture feature extraction for texture classification. Texture segmentation involves the partition of the image into region of the uniform texture. To do so, it is essential to find a set of texture features having good discrimination power. This can be done by multiresolution decomposition using wavelet transformation. So, the segmentation technique makes a hierarchical decomposition of the image, and then makes the texture segmentation. We propose in this paper to show how wavelet decomposition combine with MRA can be better way for texture segmentation.
3. MULTISCALE IMAGE ANALYSIS
The drawback of Statistical method of analyzing images is not based on the image roughness or texture of an image. We utilize a recent development of wavelet transform for image processing application to explore edge-based segmentation and texture based segmentation. The procedure is as follows: Take any image to be analyzied, convolute it with Quincunx[9][11] matrix (derived of Law's texture mask) or Geronimo, Hardin, and Massopust (GHM) filters[5] [3] at several levels of resolution (7 levels). The GHM are:
Each coarser level is obtained by shrinking the image approximately by a factor of one half. Then, at each level, the image is interpolated back to its original size resulting in a blurred image. We then proceed to construct slope image as follows: Apply a sparce wavelet mask or capacity to each level l at each pixel point p to get a coefficient clp and compute the roughness coefficient or scaling exponent p which is the slope of the least square fit log2clp versus log22l. Partition the range values into n (equal) parts and segment the image accordingly. Next, form a black and white (binary) image for each segment represents the characteristic function of each segment. To complete the segmentation process, extract original image for each characterized slope. An alternate method would be to start with seven interpolated images and apply any symmetric edge mask, then repeat the above procedure.
4. RESULTS AND DISCUSSIONS
To illustrate out method, a erial photographic images on Pleistocene-age glacial deposits in west-central Ohio were used in the study. The area was selected because of the ancillary data that was available for verification of the classification being experimented with. In this study, all the aerial photographic images were captured and digitized at 100 dpi using a Microtek Scanmaker X6 scanner. The imagery is 1981 infra-red aerial photography that is provided by the United States Cartographic and Geospatial Center in Fort Worth, Texas for the use of soil survey work. The scanned color imagery was then converted to grayscale for analytical purposes. The original image is shown Figure 4(a).
Figure 4 (a)
Traditionally, each photographic image is classified into the four major land cover classes (cultural areas, hydrographic areas, vegetative areas, and bare soil surface areas). Currently, soil scientists for the Ohio department of Natural Resource Conservation Service compile these maps manually by delineating soil boundaries using pens and/or pencils. The delineation boundaries are drawn in the field and then transferred onto a mylar photobase and overlays. Using the multiscale image analysis method the field boundaries can be detected using edge detection. Edge detection algorithms[6] were found only to be effective in some areas and would leave gaps that required manual editing. This is not viable on a large area, which can be done by using texture classification. It was established that the most effective way to locate fields was through the identification of edges and textures by using the wavelet method.
Figure 4 (b)
Figure 4 (c)
Texture feature images obtained by convoluting the original image with multiscale wavelet coefficients are shown in figure 4(b)-figure 4(c). We then construct the slope image S as explained in section 3. This step generates features in which texture information is enhanced. The slope image S can be segmented by using thresholding techniques. Figure 4(d)- Figure 4(f) shows three segments of the image.
Figure 4 (d)
Figure 4 (e)
Figure 4 (e)
Further, the wavelet method will not remove important detail from the image whereas the traditional method does. The use of this method helps us to measure the degree of roughness of the soil image and classify it into different classes. That is, we are segmenting the image into different features including soils. Our experimental results show that this method, would be much faster than the traditional manual method to classify the soil images. This evaluation suggests that this model is an excellent tool for analyzing the surficial soil images
5. CONCLUSION
In this paper, we presented a new multiscale wavelet based texture feature extraction method to classify and characterize surficial soil images. This evaluation suggests that the model is an excellent tool in analyzing the surficial soil images. However, additional refinement of the multiscale analysis capabilities needs to be explored. This would be performed by electronically scripting a program that would group the pixels with similar qualities and to delineate these areas of similar pixel qualities. These delineations would then be tested and correlated in the field or by overlaying an electronic map of the field-verified soil delineation boundaries for comparison purposes. In addition, this method has the potential for soil scientists to utilize historical aerial photographs as additional analytical tools. Further studies are needed to examine this finding in more details.
6. REFERENCES
[1] G. Strang and T. Nguyen, Wavelets and filter banks. Wellesley, MA: Wellesley-Cambridge, 1996
[2] Couloinger et al., Benefit of the future SPOT 5 and data fusion to urban mapping. International Journal of Remote Sensing, 1998
[3] X. Xia, et al., Design of prefilters for discrete multiwavelet transform, IEEE Trans. Signal Processing, vol 44, no. 1, Jan. 1996.
[4] Pichler, Oalaf, Comparison of texture feature extraction using adaptive Gabor filtering, pyramidal and tree structured wavelet transform. Pattern Recognition vol. 29, p 773-742 1996
[5] Reza Nazafat, Hamid, Multiwavelet based feature extraction of MRI images, SPIE., Wavelet application in signal and image processing, Volume 3458, 1998
[6] Soundararajan, Rick A. Robbins, Robert A. Hovis Analysis of Soil Properties and Characteristics Using Rescaled Range(R/S) Method, EUROIMAGE,2001 International Conference on Augumented, Virtual Environments and 3D Imaging, MyKonos, Greece, May 30- June 1, 2001,
[7] Y.Meyer. Ondeletted et operatrurs, Tome 1, Hermann Ed., 1990
[8] S. Mallet. A theory for multiresolution signal decomposition: the wavelet representation. IEEE Pattern Anal. And Machine Intell., 11, 1989.
[9] D.J.Hebert Soundararajan E., Wavelet, Fractal and Multifractal image Analysis & Compression, International Journal of Computers and Applications, 2000 preprint.
[10] D.J.Hebert, Soundararajan E., Fast Fractal Image Compression with Triangulation Wavelets, Proceeding SPIE Conference on Wavelet Applications in Signal and Image Processing VI, Volume 3458, 150-159, 1998.
[11] Soundararajan Ezekiel, James M. Kramer, Robert A Hovis, Seismic Signal Processing by using Rescaled Range (R/S) Analysis Based Fractal Dimension, Proceedings of the IASTED International Symposia, Innsbruck, Austria, Feb 19-22,541-544, 2001.