17-03-2017, 01:59 PM
Image recovery is an important issue in the field of pattern recognition and artificial intelligence. Searching or retrieving images based on their content is called content-based image retrieval. In CBIR, images are indexed by their visual content such as colour, texture, shapes. There are several methods implemented and all these methods are also compatible with the user feedback system to fine tune the search results. But those methods are not practical for real applications. The new framework method, the Navigation Pattern-Based Relevance Review (NPRF) is used to achieve high efficiency and effectiveness of the CBIR. In terms of effectiveness the proposed search algorithm NPRF Search makes use of discovered navigation patterns and three types of query refinement strategies, Query Point Movement (QPM), Query Reweighting (QR), and Query Expansion (QEX) to converge The search space Towards the intention of the user effectively. By using NPRF method, high-quality RF image retrieval can be achieved in a small number of feedback.
Growing multimedia demand has increased the interest of multimedia content mining recently. The recovery of images from a large multimedia repository is a difficult but interesting task. Due to the complexity of multimedia content, understanding the image is difficult. Extracting valuable knowledge from a large-scale multimedia repository, so-called multimedia mining is also difficult. Normally, in the development of an image request system, the retrieval of semantic images depends to a large extent on related sub-titles, for example, filenames, categories, annotated keywords and other manual descriptions. Unfortunately, this kind of text-based image recovery always suffers from problems. These include the problem of visual diversity and the problem of exploration convergence. Existing methods refine the query over and over again by analyzing specific relevant images collected by users. Especially for composite and complex images, users can go through a long series of feedback to obtain the desired images using current RF approaches. To solve the above mentioned problems, a new method called Navigation-Pattern-based Pertinence Feedback (NPRF) is used to achieve the high quality of RF CBIR recovery using the discovered navigation patterns. In terms of efficiency, navigation patterns extracted from the user's query log can be viewed as the shortest routes to the user's interested space. According to the patterns discovered, users can obtain a set of relevant images in a process of refining online queries. Therefore, the problem of redundant navigation is successfully solved. In terms of effectiveness, the proposed navigation pattern search algorithm (NPRFSearch) merges three query refinement strategies, including Query Point Movement (QPM), Query Reweighting (QR), and Query Expansion (QEX) to address the problem of Exploration convergence.
To achieve the high efficiency and effectiveness of CBIR for that, using two types of methods for extracting features such as SVM (support vector machine) and NPRF (navigation-relevant relevancy feedback). Using the svm classifier as a category predictor of query and database images, they are initially exploited to filter irrelevant images by their different low-level, concept and key-point characteristics. Therefore, we can reduce the query query size in the db, then we can apply NPRF algorithm and refinement strategies for additional extraction, then we can combine the colour and texture of the given query image to get the optimal solution .