Improving the Performance of Ontology Based Image Retrieval using SPARQL Query Language and Decision Tree Algorithm

Transcription

1 Volume 1, No. 11, January 2013 ISSN The International Journal of Computer Science & Applications (TIJCSA) RESEARCH PAPER Available Online at Improving the Performance of Ontology Based Image Retrieval using SPARQL Query Language and Decision Tree Algorithm N.MAGESH M.E., Assistant Professor, Dept of Computer Science and Engineering, Institute of Road and Transport Technology, Erode Tamil Nadu, India nallasamymagesh@gmail.com DR.P.THANGARAJ Ph.D., Professor, Dept of Computer Science and Engineering, Bannariamman Institute of Technology, Sathyamangalam ABSTRACT Semantic web uses ontology structure for annotating images. The ontology based framework is developed for representing natural image domain. The textual and visual features of images are extracted and stored as the part of the ontology. The ontology is represented in OWL (Web Ontology Language) format which is based on RDF (Resource Description Framework) and RDFS (Resource Description Framework Schema). Internally, the RDF statements represent an RDF graph which provides the way to represent the image data in a semantic manner. Various methods are used to retrieve the semantically relevant textual data from ontology model. The SPARQL query language is more popular methods to retrieve the textual data stored in the ontology. The textual or keyword based search is not adequate for retrieving images. The end users are not able to express the visual features of an image in SPARQL query form. Moreover, the SPARQL query provides more accurate results by traversing through RDF [18] graph. The same types of images are retrieved by providing the one to one mapping between extracted features and stored features. The other relevant images can not be retrieved by one to one mapping. So the relevancy can be provided by some kind of onto mapping. The relevancy is achieved by using a decision tree algorithm. This paper proposes a method to enhance the semantic image retrieval by using the SPARQL query language and decision tree algorithm along with the false filtering method. KEYWORDS: Ontology, RDF, RDFS, OWL, SPARQL, Inference, Semantic web, XML, Image indexing, ontology, decision tree learning. 1. Introduction 2012, - TIJCSA All Rights Reserved 79

2 The present World Wide Web (WWW) stores the huge amount of data and the capacity of data is increasing in an exponential manner. Searching the useful information from the huge data is a tedious task in the current web which is mostly depends on keyword based search. The vector space model is used to retrieve the data stored in the web. The performance of the system depends on matching the keyword within the available data. Such a model does not consider the semantic information present in the text. Current search engines use an inverted index method for indexing a particular web page. The textual information about a particular web page is extracted by crawler. They are simply keywords. Keyword based search lacks in meaning and relationships among data and thus machines are not able to understand the data [3]. 1.1 Drawbacks of Today s Web The web contents are stored in web pages in an unstructured way. The search engine retrieves the web pages using the degree of relevance. Most of the results produced by the search engine are irrelevant to the user. The computers are not able to understand the data maintained on the web. In Google and Yahoo search engines [12], image search is still a problem due to lack of meaning and relationship of the images stored in the web. Presently, the image is retrieved by using the keyword based search. Therefore, effective search and retrieval methods are needed for image extraction in the semantic web. The above drawbacks are eliminated by the semantic web. 1.2 Methods to overcome the Drawbacks The search engines simply perform the string comparison operation to find a keyword. It does not have any knowledge to retrieve and display more relevant data to the user. The web pages are displayed by using the degree of relevance and do not consider the semantic relationship among data. The image search and retrieval is still a great difficulty in the present web. These problems are rectified when using an ontology structure for the representing the images. This paper proposes the method to improve the performance of image retrieval by combining SPARQL query language and decision tree algorithm [10] and evaluate the image retrieval performance using precision-recall graph [19]. 1.3 Problem Statement An image ontological database is created for the natural image domain. The ontology stores the images and its features in a semantic manner. If the visual features are directly given to the SPARQL query as input, this will retrieve the images available in ontological structure. But the users are not aware of visual features of an image, the decision tree algorithm is used for learning the visual features of the input query image. The objective of this paper is to propose the ontology based image retrieval using SPARQL query and a decision tree algorithm [10] In Section 2, ontology based image retrieval method is explained. Sections 3 describe about the image ontology construction for a general image domain. Section 4 explains the overview of semantic search, the usage of SPARQL in image retrieval using high level features and the usage of the decision tree method in image retrieval using low level features. Section 5 discuss about false filtering algorithm and its concepts. Section 6 is the test environment and comparative analysis of the SPARQL query, decision tree method and combined approach, Section 7 gives the conclusion and future work. 2. Ontology Based Image Retrieval (OBIR) 2012, - TIJCSA All Rights Reserved 80

3 Ontology is a knowledge representation method. It uses classes and properties for organizing the knowledge. The images and its features are represented as instances of the classes. The image ontology is constructed for the image domain [12] which includes text and visual features along with relationship among data. In the semantic web, the image is represented as instance of classes. The objects present in the image is extracted and these are represented as keywords. It is called as keyword based indexing. The high level features of an image are represented as keywords like Name, Creator, Date, Location etc. The user is allowed to search the image by selecting one, or more, keywords. The visual retrieval (low level) involves retrieval of basic features such as color, texture, shape and object location. CBIR (Content based Image Retrieval) is an efficient technique when considering visual features of the image and TBIR (Text Based Image Retrieval) is efficient when considering high level data. Both representations do not consider the gap present in human perception of an image. This gap is called as semantic gap which remains as a challenge to be overcome. One possible way to overcome this problem is to represent the image by using the combinations of TBIR and CBIR along with the relationship between the data. This can be done by using ontology structure [15]. 2.1 Ontology Reduces Semantic Gap The ontology considers the semantic relationship present in the image by using the predicate. For example, the word red rose is different from red ball or red circle. The red rose semantically implies the flower rose whose color is red. It is represented by using the relation ( predicate ) has_color. The RDF syntax is Rose. has_color. Red The relation has_color is also used to distinguish the rose containing red color from other roses whose color may be white. Similarly the word happy person is different from the person. The RDF syntax is Person. has_mode. Happy The relation has_mode is used to distinguish the Person having happiness from other persons who may be sad or normal. There is still a large semantic gap existing between the visual level and high level image representation. The actual semantic meaning is identified by using inference on its parent classes. This can be achieved by means of an ontology. Since the ontology maintains background knowledge as classes and relationship, inference is taken from the existing facts. The reasoning tools such as Pallet, Fact are used in taking inferences. However, the fundamental unsolved problem in CBIR is the semantic gap that is, the missing link between the semantic categories that a user is looking for and the low-level features that CBIR systems offer. The ontology based image representation is used to reduce the semantic gap exist between visual and high level image representation and used to retrieve images whose content is complicated and partly unknown to the user [6]. 3. The Image Ontology Construction Image ontology is constructed using class, properties and instances. A class hierarchy is created using a tree structure as shown in figure 1. The PROTÉGÉ is used in building a perfect ontological hierarchy and can be used to represent exact interconnection relationships between the classes using properties. Ontology is created for the natural image domain. The ontology 2012, - TIJCSA All Rights Reserved 81

4 based semantic annotation manages three main elements, namely, ontologies (metadata), documents (image data, or content in unstructured form) and annotations (links between the data and the metadata). The image ontology has two components, namely, the class hierarchy of a domain and the textual description of the domain. The textual is further divided into text description and visual text description. The ontology model is constructed for providing shared semantic interpretation of image contents. This is used to keep all the necessary information about the images in the ontological database [2] [6]. 3.1 Image Annotation Figure 1. Structure of class and instances Figure 2 shows the annotation of the elephant image in the ontological structure. The labels are used for denoting additional information such as high level and visual features of the image. An image and its values are annotated to one or more class by using slots and labels. The class hierarchy enriches the annotations. The image with the same category is easily identified and retrieved by using the features [1]. 2012, - TIJCSA All Rights Reserved 82

5 Figure 2. Linking image with ontologies (Image Annotation) The mathematical operations like union, intersection, cardinality etc are applied in the ontology. The semantic image indexing [15] is done by annotating the image to the class [1]. 4. Semantic Search The semantic search is different from ordinary keyword based search. The semantic search [11] consists of the construction of a query engine that receives requests in an ontology query language (such as SPARQL), executes them on the ontological structure and returns tuples of ontological values that satisfies the conditions in the query. This method is a Boolean search on RDF graph. Since it is a formal ontological structure, it avoids ambiguity and redundancy. The semantic meaning is identified by performing backtracking on the classes. The reasoner is used in checking the consistency of data in an ontology and performs semantic reasoning on the ontology. The ontology query languages allow expressions to be written that can be evaluated against ontology [4]. The existing query languages are OntoQL, SPARQL, SeRQL, TRIPLE, RDQL, N3, and Versa. The SPARQL query language has been adopted by the W3C (World Wide Web Consortium) as the means to query ontologies built using RDF and has been extended to support OWL format [16]. 4.2 Searching for Image The ontology contains the image and its annotation values in low, intermediate and high levels. The RDF/XML codes are automatically generated by the protégé. Internally, the codes represent the RDF graph pattern [19] which is compared against query in the information retrieval. The SPARQL is used to perform text based pattern search and able to retrieve intermediate and high level data. The visual features are not compared by using a text based search process. Moreover the users don t have the knowledge of visual feature. So an alternate mechanism is required for processing visual features. This can be achieved by using a decision tree algorithm [14] [18]. The ontological image database is created with low level and visual features as annotation values. The owl file stores the image and its features in RDF form which represent the RDF 2012, - TIJCSA All Rights Reserved 83

6 graph. The similar images from the owl file are retrieved by using two methods. The first method uses SPARQL query and the second method uses the decision tree algorithm. SEMANTIC IMAGE RETRIEVAL HIGH LEVEL RETRIEVAL USING SPARQL QUERY LANGUAGE LOW LEVEL RETRIEVAL USING SPARQL QUERY LANGUAGE AND DECISION TREE ALGORITHM Figure 3. Types of queries for the Image Retrieval Semantic Image Retrieval using SPARQL query language The SPARQL is a query language for ontologies. Semantic searching is done with the help of the SPARQL query language. SPARQL is an RDF query language which is able to retrieve and manipulate data stored in RDF format. The SPARQL contains capabilities for querying graph patterns along with their conjunctions and disjunctions [15]. The query is specified in the triple pattern which is subject, object and predicate. This considers the semantic relationship exist among the data. A view based search is recommended for image manipulation. The user interface is designed for getting query word as input. It is converted into SPARQL format and applied on the image repository. The RDF pattern match is performed on ontology. Finally the system returns the image references relevant to the query format. These images are presented to the user after ranking. It is very difficult to write the SPARQL query for expressing visual features of an image. The following queries are used to extract images using high level features stored in image ontology [13] [9]. Query 1. Answer Find Subject, Predicate and Object nodes. SPARQL FORM: SELECT?S?P?O WHERE {?S?P?O. } LIMIT 4 Eagle rdfs:subclassof Bird Sunrise rdfs:subclassof Flower SimpleInstance Image_10.jpg rdf:type Swan SimpleInstance Image_12.jpg Theme Elephents 2012, - TIJCSA All Rights Reserved 84

7 Query 2. Answer Query 3. Answer Find sub classes of bird. SPARQL FORM: SELECT?x WHERE {?x rdfs:subclassof :Bird } Eagle Peacock Swan Find the picture of birds with white color SPARQL FORM: SELECT?y?x WHERE {?x rdfs:subclassof :Bird.?y :Color "White" } Swan2 Swan Duck Query 4. Answer Query 5. Answer Find the creators of Swan Images. SPARQL FORM: SELECT DISTINCT?name WHERE {?x rdf:type :Swan.?x :Creator?name } Mr N.Magesh Mr M.Aswinth Find the picture of Car with red color SPARQL FORM: SELECT?x?y WHERE {?x rdf:type :Car.?x :Color "Red" } Car1 Car2 Car Semantic Image Retrieval using Decision Tree Method Figure. 4 shows the proposed semantic image retrieval system using decision tree (DT) learning method. The visual features of the images are represented by using statistical properties. The image invariant features are taken for analysis. The features like color and texture are taken for analysis. First, a set of sample attributes is collected for the 20 frequently used concepts as training data. The image is divided into a number of segments using a JSEG image segmentation algorithm [7] [11]. 2012, - TIJCSA All Rights Reserved 85

8 Figure 4. Semantic Image Retrieval System The color and texture features of each image region are extracted by using image processing tools and stored in the feature vector. The decision tree is trained to remember the set of values. Secondly, the query image is given and it is also trained to the existing values. The training generates tree which is the decision tree classifier. The low level features are stored in a feature vector. The decision tree is trained by using the low level features. The topmost node in the tree contains the values whose information gain is more. Thirdly, the few classifiers starting from the root are taken. The decision tree testing is used to compare the low level features of images stored in the ontology with the decision tree classifier. The SPARQL query is generated using the classifiers as input. The images are retrieved from ontology. The matching produces the relevant images and eliminates the irrelevant images. Finally, the relevant images are displayed after ranking. The search efficiency is improved by using minimum depth and breadth of the tree. The operations such as data discretization, pre pruning and post pruning are used to improve the efficiency of decision tree learning [8]. 5 False Filtering Algorithms A simple algorithm named False Filtering ( FF) is used to remove the irrelevant images in the initial stage. The content based image retrieval using DT takes more time for comparisons. Approximately it takes 70 comparisons for each segment of an image. If the image is having five segmented regions, then the search needs 5 * 70 = 350 comparisons. This is can be reduced by false filtering algorithm. The text feature of the image is taken and it is indexed to a hash table as shown in Figure Concept 2012, - TIJCSA All Rights Reserved 86

9 A well-known color vision theory is the opponent color theory, which suggests that there are three visual pathways in the human color vision system. One pathway is sensitive mainly to light dark variations, this pathway have the best spatial resolution. The other two pathways are sensitive to red green and blue yellow variation (the opponent channels). The blue yellow pathway has the worst spatial resolution. In opponent-color representation, the spatial sharpness Figure 5. Color appearances in the red green (rg) and yellow blue (yb) chromatic space (the light dark component was fixed to a constant). of a color image depends mainly on the sharpness of the light dark component of the images and very little on the structure of the opponent-color image components. There is evidence to suggest that different visual pathways process color and spatial pattern in the human visual system. These human vision theories suggest that the colors of an image are mainly encoded in the opponent color signals whilst the spatial patterns or textures are mostly encoded in the light dark component. In the current work, the usage of colors is concentrated and therefore only uses the opponent color components to index the images. The texture is added by using the light dark component of the image follow a similar principle. Assuming that the original data is in RGB color space, the red green (rg) and blue yellow (by) components can be calculated as rg = R G, by = ½ (R + G) B ( 1 ) Figure. 5 plots the chromaticity diagram in the rg by space. As we move from the origin along the rg-axis the colors changes from green to red, whilst moving from the origin along the by-axis the colors change green/red to blue. Clearly, a certain value of rg or by has an intuitive correspondence with the color appearance, e.g., a large rg value indicate the red color and a large by value indicates the blue color, etc [19]. Table -1. The Color median and Texture values of images Color Texture S.No Image R M G M B M Entropy Energy 2012, - TIJCSA All Rights Reserved 87

10 Our idea is to directly use the chromaticity diagram and use the rg and by signals as the indexing features. Some statistical value are obtained by dividing the image database into clusters, each having a particular visual theme corresponding to a color, or equivalently, a particular pair of rg/by values on the chromaticity diagram. In this way, a simple and intuitive link amongst the (color) appearance of the image is established; the numerical representation of the image is given by using the value. The rg/by opponent components values are used to eliminate the irrelevant images. 5.2 Hash Table The hash table contains the RGB rg and by color and texture value of the image name to be searched. First the image name is taken for the analysis of the image. For the given image name, the index values are accessed using hash table as shown in Figure 6. The rg and by values of images stored in the ontological structure is compared with the entries stored in the hash structure. If the comparison is successful, then the image is compared with decision tree. The initial test is used to eliminate the irrelevant image using minimum number of comparisons. Mainly it is used to remove the unwanted images with minimum number of comparisons at the beginning. It is called as feature map hashing [8]. 2012, - TIJCSA All Rights Reserved 88

11 Image Lion Elephant Tiger Swan Sparrow Rock Index RGB rg >= 16 & rg <= 36 by >= 36 & by <=75 Texture Entropy =12.34 Energy =.006 Null 3 4 RGB rg >= 8 & rg <= 57 by >= 23 & by <=77 Texture Entropy = Energy = Null 5 RGB rg >= -1 & rg <= 46 by >= -4 & by <=57 Texture Entropy =11.37 Energy =.0050 Null Figure 6. False Filtering Hash Structure for color and texture checking Each query in our experiments corresponds to a high-level concept. It is assumed that each image has at least one dominant region representing its semantic meaning. Given a user query, for each image in the returned list, if any region it content is relevant to the query concept, then the image is considered relevant to the query. If none of the region it contains is relevant to the query, then this image is considered irrelevant to the query. In this way, we can filter out those images which are irrelevant to the query from the list. Consider the simple linked hash queue structure. The various images and its index values are stored as shown in figure. 6. The linked list stores the primary components of an image, the component may be color component or texture component. The size can be extended further. The image features are used to find images in the database which are the most similar to the query image. A drawback, however, is that these features cannot adequately represent the image semantics. Extensive experiments on CBIR systems show that low-level content descriptors often fail to describe the high level semantic concepts familiar to users. This algorithm is more suitable for the images which are not having more sub objects. If the image is having more sub objects, then multiple accesses to hash table will solve this problem. The color median and Entropy, Energy values are calculated for 10 images in each category, The rg/by approximation is represented in the hash table. It is used to eliminate the irrelevant image from the query input as keyword. 6. Test Environment and Result Analysis In information retrieval, a perfect Precision score of 1.0 means that every result retrieved by a search was relevant whereas a perfect Recall score of 1.0 means that all relevant documents were retrieved by the search. 2012, - TIJCSA All Rights Reserved 89

12 Precision In the field of information retrieval, precision is the fraction of retrieved documents that are relevant to the search Number of retrieved images relevant to the query image precision = Total number of images retrieved (2) Recall Recall in Information Retrieval is the fraction of the documents that are relevant to the query that are successfully retrieved. Number of retrieved images relevant tothe query image recall = (3) Total number of relevant images inthe database 6.2 Querying using SPARQL query Since the SPARQL query does not take the image as input, the user interface is designed to get the image. The Figure 7 shows the workflow for image retrieval system using high level features. The image ontology constructed for 820 images. It has 56 concepts (classes and properties) and 19 low level features. Number of total relevant documents in this collection is counted as 23. The decision tree was trained by using 20 general images. Using Eq. (2) and Eq. (3), the precision and recall values for the query image are calculated for the proposed method [12]. The values in Table II represent the Precision and Recall value when using the SPARQL query for text as input. Figure 7. Query by using User Interface Table II - The precision and Recall value - SPARQL query for text as input 2012, - TIJCSA All Rights Reserved 90

13 The comparison of precision and recall of the retrieval system with SPARQL query and decision tree algorithm is given below. The decision tree gives the best output if the image invariant parameters are taken for training. The performance of the proposed method can be identified by using precision and recall. Specifically the keywords Red Car and White Bird are identified by the SPARQL using semantic relationship. 6.2 Querying using Decision Tree The unknown image is given as input to the system. The structure of decision tree changed every time if a new image is given. The low level features are extracted and stored in the feature vector. The decision tree is trained by using the normalized data. Figure 8. Assigning Input to the Proposed Method 2012, - TIJCSA All Rights Reserved 91

14 Finally the decision tree output is given to SPARQL query. The images matching with the decision tree are only taken by means of SPARQL query. The values in Table III represent the precision and Recall value when using the decision tree algorithm for query image as input. Table III - The precision and Recall value - Decision tree for query image as input Both precision and recall are therefore based on an understanding and measure of relevance. The keyword based retrieval has achieved the precision as 100% when using the SPARQL query. The decision tree based retrieval has achieved the precision as 80% as shown in Figure 9.b. 9.a SPARQL query using high level features 9.b Decision tree algorithm using Low level features 2012, - TIJCSA All Rights Reserved 92

15 Figure 9. Comparison of Precision-Recall 6.3 Integration of SPARQL and Decision Tree using Fast Filtering algorithm The text-based image search and the semantics-based decision tree system are integrated to improve the performance of a semantic image search system. The irrelevant images are removed by means of an algorithm named False Filtering (FF) which is based on opponent color theory. Table IV. The precision and Recall value Text and Image as input Input Query Retrieved Documents Total Retrieved Image Text Documents Relevant Irrelevant Precision Recall Swan Red Car Giraffe Tiger Rose Eagle A few practical ways are proposed by which FF can be hierarchically integrated with contemporary text-based search engines [19]. In all scenarios, text-based image search results for a given query are obtained first. Then, FF is used to filter out those images which are semantically irrelevant to the query. The efficiency is further improved by the SPARQL query when using text as the part of the image. The precision value has been improved by using ontology because it eliminates the irrelevant images by well defined relationship. The combined approach has achieved the precision as 85 %. The comparison of precision and recall of the retrieval system with SPARQL query and decision tree algorithm is given below. The performance of this algorithm is fully based on the decision tree. The decision tree gives the best output if the image invariant parameters are taken for training. 2012, - TIJCSA All Rights Reserved 93

16 0.9 Decision Tree - Graph Precision Recall Figure 10. Comparison of Precision-Recall 7. Conclusions This paper proposes the semantic image retrieval system using SPARQL query and a decision tree algorithm. The domain specific image ontology is created. The high level, semantic based relationships and visual features are taken for analysis. The semantic and high level features are handled by means of the SPARQL query language. The visual features are handled by using a decision tree algorithm. The decision tree algorithm is used to classify the image data in perfect manner. As for as the decision tree algorithm is concerned for visual features, there is improvement in precision and some increase in recall values. The combined approach is used to retrieve the relevant images in an accurate manner. The implementation results illustrate, that this type of image retrieval, effectively retrieves the images that are very close to the query image from the ontological database, this could be visualized from the precision and recall plots determined from the retrieval results. The proposed work can be extended to include multimedia data such as Sound, Video and Music. The filtering method can be improved to get high accuracy. The decision tree algorithm can be extended to the handling of missing data. If the histogram is used for color representation, the performance can be improved. References 1. N.Magesh, Dr. P.Thangaraj, Semantic Image Retrieval Based on Ontology and SPARQL Query, IJCA (International Journal of Computer Applications), N.Magesh, Dr. P.Thangaraj, An Image Retrieval System Based on Extensive Feature Set Using ID3 Decision Tree Algorithm, European Journal of Scientific Research, ISSN X Vol. 89 No 1 October, 2012, pp D.fensel. The semantic web and its languages. IEEE Intelligence Systems, The semantic web primer book by Grigoris Antoniou, Frank Van Harmelen. 2012, - TIJCSA All Rights Reserved 94

17 5. David Picard, Matt Cord and Arnaud Revel, "Image Retrieval over Networks: Active Learning using Ant Algorithm," IEEE Trans on Multimedia, Vol.10, No. 7, pp , Chih-Fong Tsai and Chihli Hung, "Automatically Annotating Images with Keywords: A Review of Image Annotation Systems," In Proc. of Recent Patents on Comp Sci, Vol. 1, pp , D. N. F. Awang Iskandar, James A. Thom and S. M. M. Tahaghoghi, "Content-based Image Retrieval Using Image Regions as Query Examples," In Proc. of the nineteenth conference on Australasian database, Vol. 75, Australia, Ying Liu, et all, "Region-based image retrieval with high-level semantics using decision tree learning," Journal of Pattern Recognition, Vol. 41, No. 8, pp , Aug Mario Arias Gallego, Javier D Fernández, Miguel A Martínez-prieto and Pablo De Fuente, "An Empirical Study of Real-World SPARQL Queries," Journal on challenges, Vol. 33, No. 3, pp. 3-6, D.V.N Harish, Y. Srinivas, K.N.V.S.S.K Rajesh and P. Anuradha, "Image Annotations using Machine Learning and Features of ID3 Algorithm," International Journal of Computer Applications, Vol. 25, No. 5, pp , Jul M. Koskela, J. Laaksonen, S. Laakso, and E. Oja, The picsom retrieval system: Description and evaluations, In Proc. of The Challenge of Image Retrieval, Brighton, UK, May B. Ramamurthy and K.R. Chandran, "Content based Image Retrieval for Medical Images using Canny Edge Detection Algorithm," International Journal of Computer Applications, Vol. 17, No. 6, pp , Mar Syed Abdullah Fadzli, Rossi Setchi, "A Semantic Approach To Text based Image Retrieval Using A Lexical Ontology," In Proc. of International Conference On Kansei Engineering And Emotion Research, Keer, Paris, Mar Jun Zhai, Kaitao Zhou, Semantic Retrieval for Sports Information Based on Ontology and SPARQL,International Conference of Information Science and Management Engineering, Soner Kara, Ozgur Alan, Orkunt Sabuncu, Samet Akpınar, Nihan K. Cicekli, Ferda N. Alpaslan, An Ontology-Based Retrieval System Using Semantic Indexing, IEEE, ICDE workshop J. Huang, D. J. Abadi, and K. Ren. Scalable SPARQL querying of large RDF graphs. In VLDB, Jens Lehmann, Lorenz Buhmann, Auto SPARQL : Let Users Query Your Knowledge Base, Guoping Qiu, Jeremy Morris, Xunli Fan Visual guided navigation for image retrieval, Science Direct, Pattern Recognition, 40, , , - TIJCSA All Rights Reserved 95