SOFT COMPUTATION BASED INDIAN TOPOGRAPHIC MAP UNDERSTANDING SYSTEM (ITMUS)

Abstract

A topographic map is an important source of information for geospatial planning and analysis. Apart from conventional uses, nowadays topographic maps have been extensively used for the development of automated Geo-information system including GIS and decision support system. These mostly rely on the topographic map to obtain geographic information, location details, the extent of urban area and landscape analysis, etc. The information of interest to any geospatial application requires being extracted from topographic maps. At present, extraction of information is being carried out through manual digitization or semiautomated methods which require expertise and human intervention. Also, these approaches are slow and error prone. Hence, there is great need to automate extraction of information from the topographic map. Few systems have been developed for the interpretation of topographic map, but the approaches are insufficient to automatically extract geospatial information and are incapable of understanding the topographic map. The research work as done so far not lead to a precise formulation of the map objects description and classification. Most of the map interpretation systems emphasize particular coherent organization of maps or raster to vector conversion process. To provide a solution for the manual digitization and to access the topographic map information efficiently, sophisticated approaches for semi-automatic selective map interpretation have been reported by many researchers. The proposed approaches yield good recognition rates for the well-isolated objects of interests. Few existing systems have dealt with the problems of recognition of limited map objects, but most of them have confined the scope of research for linear object's reconstruction and recognition. Most of the systems have not considered extraction of geospatial information. The topographic maps are quite complex. Complexity is due to color ambiguity, spatial ambiguity and pattern indiscrimination present in the topographic map. Existing approaches have not provided any robust framework for handling color information and pattern indiscriminate present in a topographic map. However, the human map reader is excellent in pattern recognition. Hence, the topographic map understanding approach must be more natural to conceive based on the humanistic approach rather than recognizing edges, object boundaries and syntactic arrangement of structure primitives. Hence, a generalized framework has been needed to handle the high variability of graphical content displayed on the topographic map. Data or information capturing by a human is quite efficient. But human understanding may get bias due to inconsistencies. In contrast, computers are more deliberate and less prone to error. Hence, the research work has been visualized by the concept of ii integration of human and computer map processing towards the automated understanding the topographic map. The main objective of this research work is to develop an automated system for the understanding of Survey of India (SOI) topographic map. To achieve the objective of topographic map understanding the human mentation and learning capabilities have been emulated by computer treatment to develop a soft computation based Indian Topographic Map understanding system (ITMUS). The first phase in human map understanding process is reading of map legends which uses color and geometrical appearance of legend. The Legend understanding subsystem (LUS) has been devised to represent the legend numerically and described using a set of shape and structure parameters as well as color which has been used at a different level of matching. The LUS performs static rule based matching and legend recognition, which is consisting of structure and shape parameters in premise part and semantic description of the legend in consequent part in a hierarchical manner. The LUS has been tested on legend set developed by Survey of India and obtained an average accuracy of 88.424%. The interpreted legend description data set has been partitioned into training set libraries based on the data model prepared by Modern Cartographic Center, Survey of India. The second phase in human map reading is to utilize the legend description data for interpreting the objects/symbols present on a topographic map. This phase has been emulated by the Map understanding subsystem (MUS) which uses legend description data to acquire information about the legend. However, topographic map treatment by the computer system is not easy due to high density and overlapping of map objects. This situation has been dealt with, a “peeling onion strategy”, where the continuous subtraction of the already recognized map layers has been carried out using image processing techniques to simplify the processing of the rest of the complex layers. For map object interpretation, "correlation theory of brain" has been employed. It has been done by designing Fuzzy Inference System (FIS) to infer rules from initial training set libraries. Once the initial membership functions have been created, the training of system has been carried out by providing legend structure data and membership function created by FIS. After the training of system has been finished, the final membership functions and training error have been produced. The checking data have been used along with training data to make a system understand and interpret the topographic map objects. The output of ITMUS effectively conveys the interpreted information to end user. A map understanding which has been derived from MUS is represented in the form of an interpreted map (i-map) along with geo-location information. The i-map representation deals with presenting the annotated/labeled topographic map objects inside the ROI which has been selected by the end user. The ITMUS generates thematic map information in .tiff, .xlsx, .txt iii and .tab form. The thematic map layer contains geometrical and geospatial information. The ITMUS stores interpretation results of legends set in excel worksheet and in .pdf file format. The map analysis report includes the semantic meaning of the interpreting object, it's geometrical feature values, pixel location, pixel list, and geo-coordinates. ITMUS also provides insight to the user to extract the color based, feature based layers as well as intermediate description. The ITMUS has incorporated with an inbuilt accuracy assessment, a utility for both symbol and layer extraction assessments. Thus, the developed system provides a generalized framework for automated topographic map understanding by contributing automated geolocation based information extraction and retrospective map analysis for further computer-based information processing. The system performance has been evaluated by providing testing data set into the fuzzy system through the selection of a region of interest from topographic map. These data structure consists of a shape feature description of map objects present in that region. The output of ITMUS represents the semantic description code and provides a resultant understanding about the map object. The semantic code has been measured on the basis of correlations between the desired context and learning content. To evaluate and validate ITMUS with respect to different outputs, different case studies have been carried out to test different parameters. The ITMUS has been trained for various sample regions selected from the Survey of India Topographic maps having identity contains, 53C7 and 53K1 under OSM category. To carry out an evaluation of the system, different map regions have been selected from 53C7, 53F6, 53F7, 53F11, and 53K1. It has been found that the overall recognition rate of the system is 90.91%. Further, the system has been assessed on three other different criteria, i.e., its overall completeness, correctness, and rate of correct recognition. The criteria have been found to be 0.93, 0.99 and 93.79% respectively. A measure of the overall classification accuracy derived is 88.235%. The accuracy assessment and validation of ITMUS shows that system is highly robust and reliable as it is doing well for testing and checking topographic map region. Thus, it may be concluded that an ITMUS has successfully been developed. The major contribution of this research work is the integration of image processing techniques for feature extraction and reasoning capabilities of the Neuro-fuzzy model for enriching the Indian topographic information system. The major outcome of the study is the development of a generalized human based map understanding framework for automated extraction of information from the topographic map. The comprehensive conceptual formalism of map understanding and development of a robust and reliable solution for automated acquisition and extraction of information from the Indian topographic map has been made possible.