Enhancing Semantic Web Retrieval Through Ontology-Driven Feature Extraction: A Novel Proposition

: Web images represent unstructured data sets which often lead to challenges when users try to locate distinct images via text-based searches on the web. Such difficulties stem from different factors, e


Introduction
The use of web-based technologies and the internet is growing exponentially, thus web services and pages are getting heightened attention from both vendors and researchers.It is easy to see that research efforts are widely being focused on studying portals and e-commerce platforms.Nevertheless, the demand for better Quality of Service (QOS) and personalized user-centric services is also on the rise.One potential direction to address this issue is an amalgamation of semantics (i.e., creating meaningful structures) and capabilities of artificial intelligence.Therefore, our target is merging two prominent areas in the contemporary web landscape, i.e., web semantics and web mining (Casteleiro et al., 2017).Ontology plays an important role in improving our understanding of web data.Nowadays, Ontology is hugely being utilized in various research and applications domains such as including artificial intelligence, web semantics, and natural language processing (Casteleiro et al., 2017).Two implementations based on ontological representations of SNOMED CT are proposed, one based on the OWL API and the other on the W3C SPARQL 1.1 query language.The analysis of the functional capabilities required for SCTQL.
Ontology can be utilized for improving decisionmaking within the field of online media, especially images on the world wide web (Borth et al., 2013).Borth et al. (2013) provide a methodical, data-driven approach to create a large-scale sentiment ontology based on web crawling folksonomies and psychology.We often face the challenge of getting irrelevant and less precise results to user queries on search engines such as Google, Netscape, and Yahoo, which usually leads to frustrated users.Towards this problem, we turn to web images as our dataset and apply text queries in the search engine to find the exact image on the web.However, we unfortunately fall short of getting precise results.Various factors are instrumental in this problem, including redundancy in stored images; irrelevant meta tags mean false caption generation on images and incorrect associations relationships between different concepts.As we stated above, ontology plays a vital role, but constructing a viable ontology remains a big challenge as it is a very perplexing task.It is a resourceconsuming and time-intensive task.Several studies and models have been proposed to ensure the efficient reuse of existing ontologies.But the complexities persist owing to the multifaceted issues such as: 1. Locating domain-specific ontologies: Identifying domain ontologies for further reuse 2. Integrating concepts: Combining concepts from different ontologies into the desired domain by making it more coherent and comprehensive using an object-oriented approach 3. Continuous knowledge acquisition: Updating the knowledge from newly discovered ontologies 4. Effective knowledge sharing: Using the same medium to share their domain information easily This research presents our pilot attempt to build a large-scale multi-modality ontology for web image classification.The main objective is to enable the use of text-based ontology construction.Our approach attempts to harness both structural and content-related features of web data to establish real-world objects and their conceptual relationships (Chen, 2005;Liu, 2022).For future better understanding, we also train our classifiers using image features with the development of a middlelevel training set.However, our research will also show empirical results that utilize ontologies with various learning methods to enhance decision-making accuracy (Bannour and Hudelot, 2011).Image retrieval could become easy by having correct domain-specific knowledge and relationships on the web.Thus, we propose to use an agent that builds a precise before uploading an image.Next, our ontology construction procedure establishes precise relationships with existing ontology concepts based on the "an" and "is a part of" relationships.A novel procedure is introduced for easyto-use and flexible retrieval of color pictures, integrating knowledge group tools and cutting-edge image analysis methods.This technique attempts to overcome the limitations of prior frameworks, such as the limited terminology or the need to access key images and avoid physical alteration of weight.Namely, the obtained results contain a large set of static images.Utilizing ontologyconstructed text can lead us to an accurate result when the users enter a query into the search box, i.e., easy retrieval using semantics with the keyword search (Malki, 2016).Feature extraction is one of the main processes that are usually involved in creating a classification framework.The classification framework given in Fig. 1, includes image collection, image segmentation, feature extraction, image annotation, image classification on SVM, concept semantics, and image content database.
The main contribution of this study is to provide search engine users with results which closely associated with their queries.Often, the users looking for accurate results from a search engine face unrelated results due to various factors such as the most common redundancy in stored images, irrelevant meta tags, and incorrect relationships.To overcome the problem, we propose a semantic model that uses the ontology language, empowering the user to find the exact image relevant to their queries.The framework in this study ensures that a notable part of the data undergoes automated and efficient processing with a combination of the two aforementioned methodologies.Consequently, the user can get semantically enriched results aligning their query.

Ontology-Based Information Retrieval
If we take the animal class as an illustrative example, human beings can easily differentiate them using visual traits.But a machine cannot define them so straightforwardly.Machines rely on more conceptual understandings, e.g., names of animals, their diet, and their natural habitats.To attain a better level of understanding, the machine needs the use of ontology.Nonetheless, we have pertinent questions: 1. Hierarchical relationships: Is there any hierarchical relationship present on the web data that is the same as the ontology created 2. Automatic ontology construction: Can oncology creation be automated knowing that manual construction can be so time-consuming 3. Content limitation: Can we put a limit on the web content or not An ontology fundamentally shows an intricate relationship between key concepts forming a domain area.The relationship is given in Fig. 2. Upon the introduction of the ontology, we can unfold several application opportunities by merging techniques like information retrieval with web agents and semantics schemas of content.Sirea tackles the problems of content-based image retrieval and keyword-based image retrieval (Majeed et al., 2013).Ontology is very important in applications of the semantic web.Nonetheless, building image-based databases generates significant problems in finding the desired image.The semantic web is playing a very imperative role in information access by using sources like OWL, RDFs, and SPARQL (Benítez-Hidalgo et al., 2023).To address the intricacies of semantics, advanced semantic technologies using ontology is front running candidate.Like Content-Based Image Retrieval (CBIR), the integration of semantics and ontology is playing a vibrant role in image retrieval (Wang et al., 2008).This search mechanism leads to an optimized solution achieving more refined results (Negm et al., 2017;Khalid and Noah, 2011).An agent is referred to as any robot, software agent, or human being in knowledge discovery that involves the exploration of a large collection of data to identify meaningful patterns for generalized and useful conclusions.Using the OTTO text mining approach, we can extract valuable data from web applications (Bloehdorn et al., 2005).The next step is to use that data for decision-making practice using agent systems.To effectively apply decisionmaking processes, a multi-agent distributed Decision Support System (DSSMAS) logical architecture is employed, using the extracted text in the final decisionmaking process.This holistic strategy attempts to address problems in the domain, including the domain, including:  Finding domain ontologies for further reuse  Combining concepts into a desired domain by making it possible using an object-oriented approach  Acquisitive knowledge from newly discovered ontologies  Using the same medium to share their domain information The construction of ontologies by means of protégé software provides semantic understandings of image substances.In the traditional CBIR systems, the images are retrieved by using the given keyword by users.Ontology delivers a holistic approach bridging granular details of image contents, including low-level visual features, with human cognition.An ontology web language is further accustomed to form an ontology.Ontology can be constructed in two ways: Basic and domain-oriented.These ontologies have potential applications in military, geospatial imaging, and image retrieval in medical (Negm et al., 2017).A comparative analysis of different models is presented in Table 1.

Materials and Methods
In the current context, we introduce a model based on the ontological approach system architecture that defines the work of the various components of the IR system and the synergy between them.The proposed model is presented in Fig. 3.This diagram provides a comprehensive depiction of architectural complexities involved in image retrieval via semantics techniques.The semantic image retrieval system comprises the following components: Various applications use ontology which is usually hierarchy-based illustrations describing many concepts and their association or relationships.Such associations/relationships mainly manifest via word connections that are primarily categorized into three types: "Instance of", "part-of" and "is-".
In this model, an image collection methodology suited to search in large selections of heterogeneous images is presented.The proposed strategy employs a completely unsupervised segmentation algorithm to partition images into distinct areas.From these areas, low-level features encompassing color, position, shape, and size are extracted.These features are then automatically organized into suitable intermediate-level descriptors forming an easy and coherent vocabulary called object ontology.Object ontology serves as a pivotal component enabling the qualitative definition of high-level ideas that users seek in a human-centered style.During the querying procedure, irrelevant image areas are filtered out using intermediate-level descriptors.In the due course, a relevance assessment technique, using low-level features, is invoked to produce the final query results.This approach bridges the gap between keyword-based approaches, which usually assume the presence of rich image captions or manual annotation, and query-byexample approaches (Zhang et al., 2012).The model improves efficiency by proposing automated means of image retrieval (Zangeneh and McCabe, 2020).A novel hybrid semantic indexing technique is proposed with the combination of machine learning and ontology for unstructured text content (Sharma and Kumar, 2023).For the purpose to achieve higher performance and processing quantum machine learning is used effectively to compute different tasks (Umer and Sharif, 2022).Pre-trained model is used for detecting diabetic foot ulcers into suitable groups like normal/abnormal and ischaemia/nonischaemia (Amin et al., 2022).Ontology-based framework for standardization in multi-objective optimization.The framework's primary ontology, Moody, addresses anything from the formalization of evolutionary algorithms (Aldana-Martín et al., 2024).An ontologydriven reference framework (OntReF) and a Blockchainbased Reference Model (BbRM) for the SRM of both traditional and blockchain-based applications has been proposed for security purpose (Iqbal et al., 2024): <owl:Class rdf:ID="Vehicle"> <rdfs:Label>Vehicle</rdfs:Label> </owl: Class> <owl:Class rdf:ID="Vehicle Concepts"> <rdfs:Label>Vehicle Concepts</rdfs:Label> <rdfs:subClassOf rdf:Resource="#Vehicle"/> </owl: Class> <owl:Class rdf:ID="Car"> <rdfs:subClassOf rdf:Resource="#Vehicle Concepts"/> </owl: Class> <owl:Class rdf:ID="Toyota"> <rdfs:SubClassOf rdf:Resource="#Car"/></owl:Class>

Results and Discussion
In the current context, we propose a model based on the ontological approach, whose system architecture defines the work of the various components of the IR system and the synergy between them.The model's simulator is depicted in Fig. 4. The diagram provides the detailed architectural framework designed for image retrieval using semantics techniques and involves the following components used in the SIR system:  Query processing device  Matching segment  Ontology manager module

Query Processing Device
The processing device plays a vital role in receiving user input queries from the web domain.These queries indicate the particular content that has the user's interest and that they want to examine/search.The user can provide input via two methods.

Inputs Based on Text and Image
The initial technique for input in the model is purely text-based.In this procedure, the user must enter any text that contains information about the object the user wants to find.This practice is often used by existing search engines like AltaVista and Google.The motivation for integrating this scheme with SIR systems (reference model) is to offer services to users who prefer a straightforward approach without the need for intricate interactions with the model's architecture.Users are plainly required to provide textual annotations or requests, e.g., lion, fox, and elephant.A similar process is performed in the text-based query processing module (Manzoor et al., 2012).The second procedure we use for inputs is image-based (Sarwar et al., 2013).In this method, the users are encouraged to submit images containing information about the object that they seek.The input images can contain single or multiple objects and the user has the flexibility to provide descriptive choices (options) for input images.This practice is valuable when the user wants to examine the associated items/images in the same way as a conventional image search (Zhang et al., 2012).Still, this technique delivers litheness in input procedures and it offers a completely new domain to our searching algorithms.When inputs are received from the user, dedicated processing devices are utilized to tackle queries by these inputs.As a knowledge base and ontology base are used, the queries are further formulated in SPARQL language.This process contains the following two mechanisms.The example data set is presented in Fig. 5.This segment is liable for constructing the queries based on textual inputs.The process starts by detaching all regular stemmer words (-the ‖, -is‖ -and‖, -on‖…) from this input text.Next, SPARQL queries are generated with all probable-OR‖ and -AND‖.These generated queries have to pass through to matching segments.

Image-Based Query
A fundamental module is responsible for creating queries from our image-oriented inputs initiated with the perception of image objects via the shape-based feature extraction mechanism (Alani et al., 2003).After object identification, various sub-steps are executed: If the image contains the description users are looking for, it becomes a concept.After successful completion of these procedures, the SPARQL query is created using the parameters

Matching Segment
This section takes the SPARQL query as input for the query engine and works according to an ontologybased protocol to decompose the combined image.If this query returns the relevant results during the first search, the current output image will be passed to the current module for ranking the results.An Ontology example is given in Fig. 6.But if the searches are unfruitful (i.e., the relevant image is not found in our knowledge base repository), the matching module proceeds to the succeeding steps:  Image search: The preceding module searches the internet for a related image by processing the query in existing search engines like Bing or Google  Content verification: The outcomes returned by the search engines are then passed through the section of image processing for content authentication verification (Minu and Thyagharajan, 2014).

Module for Image Processing
The image changes retrieved from the search engine may enhance or diminish the user's query.Therefore, the content of any image must be validated and supported.This section deals with the submission of image-based queries aimed at image content inspection via a shapebased feature extraction process (Ryszard, 2007).If these extracted features relate to a user's query, this image will be included in the general summary, otherwise, it will be rejected.As a result, these processes ignore the irrelevant images (Banu et al., 2022).

Ontology Manager Module
This component incorporates new applicable image types, features, and notions collected from search results and then clarified and subsequently refined to improve ontological-based facts.

Module of Ranking the Images
This module can customize the image as per the user's needs.It matches images with the approved module that corrects the images to their corresponding values.The resulting database is systematically organized, which provides the same benefits.After this processing, the user sees the top ten images (i.e., multiple images are presented) and the rest of the user's queries are displayed in descending order (Manzoor et al., 2012) Fig. 7 shows the result.

Conclusion
From our perspective, the user generates a caption (text) by constructing an ontology for each image in our repository and building relationships with different concepts.This text ontology is constructed based on the information given by the text and the resultant image from the initial search.It consists of high-level content as well as low-level features like color, names, shape, etc.Then, our constructing ontology is mapped to existing domainspecific ontologies like vehicles, sports, and animals, which have already been established, based on the "an" and "is a part of" relationship.This novel approach paves the way for the flexible and easy-to-use retrieval of color images, integrating cutting-edge image analysis techniques with knowledge management tools.These methods attempt to overcome the shortcomings of approaches that often suffer from limited terms or the need for the accessibility of key images and manual input.The consequential framework is a result of extensive exploration of a large dataset of static images.After that, the given text ontology constructed can give an accurate result when users input queries in the search box.This accuracy is further improved by using semantics keyword search techniques.For more precise results, we use metadata formats, and automatic metadata generation for research papers and apply the construction of the semantic network to the multimodality ontology.Moreover, we utilize time and search optimization algorithms and search outcomes using hyperlinks, logs, and user profiles.If we include Metadata Object Description Schema (MODS), multi-object-based image retrieval, and symbolic language then it will lead to optimal results.To further improve the outcomes, modules like the Image retrieval system and semantic matchmaking module have been integrated.

Fig. 3 :
Fig. 3: Proposed model  Ontology and image analysis  System architecture query processing device  Module of matching  Image processing  Ontology manager module  The module of ranking for the image  Ontology and high-level features


Color-based extraction: Initially, the detected items are delivered to an extraction method that is based on color and practices the MTH algorithm suggested by Guang-Hai in his recent research.This step involves careful consideration of pixel color values to ensure accuracy  Structural classification: Next, the identified particles are transferred to the structural classification process so that the structure (if possible) can be identified in the detected particles  Ontological concept formation: In the third phase, the low-level profile extracted from the previous two steps is converted into a higher ontological concept.

Table 1 :
Comparison table