|
Image Processing
|
Feature Extraction In Residential Areas
By Knowledge Modelling
Semantic Knowledge Supporting Image Processing
This section concerns the verification of the previous relational structure, a hypothetical model for the analysis of a residential sub-image. A flowchart, presenting different levels of dependence and independence in the context of the sub-image is shown in Figure 2. This is a proposed early image understanding procedure, that includes four basic parts:
1. Image processing for pre-processing, data structure, and segmentation.
2. Unary independent context (or low level context) for representation.
3. Unary dependent context (or middle level context) for representation, and classification.
4. Binary dependent context (or high level context) for recognition, and image understanding.
Figure 2 A proposed IU system utilising both context and knowledge for residential areas recognition at different stages and with corresponding results.
Figure 3 Results of a test of the co-existence and adjacency of two residential objects, housing and road boundaries. The merging of housing and road patterns results in a reasonable distribution and contextual view of a typical residential area. This binary image is a result of 5% value for house reflectance and line detection. The original image scale = 1/8,000, with a resolution = 1m/pixel, image size (x:y) = 386 x 374 pixels, from Wellington, NSW.
To understand residential areas, both roofs and roads are extracted and partly identified by their descriptors. The use of road and roof maps for context considerations can lead to better identification of the objects in the residential cell. The IU system in Figure 2 shows how different levels of contextual information are processed in the study. The extracted features and recognised objects in a residential model derived from the proposed IU system is shown in Figure 3.
Concluding Remarks
A quality "primal sketch" is important for giving both global and local information image understanding. A global primal sketch (texture) provides grouping for classification purposes, while a local primal sketch (edge and line detection) provides correctly selected objects for recognition and an object-based contextual network.
The rate of house reflectance in images of residential areas for visible light is between 13% to 15%. The procedure in this study commenced with a rooftop extraction rated of 5% and varied to a rate of 12%. An intermediate rate is also recommended for extra blob information. However, for the use of only one rate, 5% is suggested, as there is a minimum of error in the extraction process, as shown in Table 1.
Low level forms of knowledge (descriptors or features), and domain knowledge (binary contextual relation) are designed for recognition methods, and have proven to be both logical and feasible. The paper has proposed a semantic network for determining residential patterns support as a photogrammetric IU system, which has been tested successfully on one example. a .
This example of residential area recognition by knowledge representation techniques based on semantic networks shows that domain knowledge is necessary to address the adjacency, co-existence, and relational structures between different objects and classes.
Acknowledgements
This research is mainly financially supported by Chung-Cheng Institute of Technology, Taiwan, Republic of China.
References
-
Ballard, D.H., Brown, C.M. 1982. Computer Vision, Prentice-Hall INC.
-
Chao, H., Trinder, J. 1997. Texture Parameter Analysis on Residential Area, 1st Trans Tasman Surveyors Conference Technical Papers, pp. 23-1/23-9, Newcastle, NSW
-
Chao, H. 1999. Road Boundary Segmentation and Representation of Residential Areas in Digital Aerial Imagery, The 18th Symposium on Science and Technology of Surveying and Mapping, pp.807/813, I-Lan, Taiwan
-
Canny, J.F. 1986. A Computational Approach to Edge Detection, PAMI Vol.8, No.4, pp. 679/698
-
Forster, B.C., Jones, C. 1988. Urban Density Monitoring Using High Resolution Spaceborne Systems, IAPRS, Comm VII, Vol.27, pp. 189/195
-
Guelch, E, Axelsson P, Stokes J, 1990. Object Description and Precise Localisation of Prescribed Objects in Digital Images. IAPRS Vol.28, part 3, pp. 221/233, ISPRS Com. III Symposium, Wuhan
-
Hough, P.V.C. 1962. A Method and Means for Recognizing Complex Patterns, U.S. Patent 3,069,654
-
Marr, D. 1982. Vision, W.H. Freeman and Company
-
Ng, T.K. 1990. Predicting Residential Housing Density and Housing Size with Satellite Imagery Using Simplified Models, Research Project Report, Master of Engineering Science, University of New South Wales, Sydney, Australia
-
Sonka, M., Hlavac, V., Boyle, R. 1993. Image Processing, Analysis and Machine Vision, Chapman and Hall Computing
-
Sowmya, A., Trinder, J. 2000. Modelling and Representation Issues in Automated Feature Extraction from Aerial and Satellite Images, ISPRS Journal of Photogrammetry and Remote Sensing 55(2000) pp.34/47
|
|
|
|
|
