A hybrid approach towards minimizing the semantic gap in image retrieval by combining classical and content based techniques

Abstract

Number of internet users has been increased drastically over past few years and so the need for secure, reliable and accurate information retrieval mechanisms. The increasing population over web creates loads of information and user data over web. It can include textual information, multimedia content, content specific URL’s, personal information and other. Due to all this retrieval of information becomes difficult on fast and efficiency scale. There are search engines that are used to retrieve multimedia data called vertical search engines. In spite of the upset in web indexes, there can be other issues which fails to meet client necessities and results in semantic gap between user search and the outcome. Other thing that can be often viewed is the content relevance. There are various search engines available that are classified into two categories known as text based search engine (traditional) and content based search engines (content based). Text-based Search Engines are simpler to implement but are sometimes contradictory in results, as they completely rely on the occurrence of query term in surrounding text. In case of absence of surrounding texts it fails to load the results or sometimes loads improper results. Also it fails to recognize the synonyms and fails to load those results. Whereas on the other hand content based retrieval is based on matching the user input query with that of the content present in the data pool. In this each feature of input query is matched against the available resources in database. Hence the results are generated with high quality and are more accurate. But it involves higher computation time and sometimes involves computation overhead. Here, in following work, a technique is used which combines both the classical approach and content based. The overview of the task being done is extracting both textual features and visual features from images and its metadata. As both techniques are combined, result will have higher content relevance and much lower overhead time and response time.