Improvement of Accuracy of Content-Based Image Retrieval Using Local and Statistical Methods

Content-based image retrieval (CBIR) system plays an important role in retrieving desired images from a large database of images. These programs in all areas, including hospitals, regulatory applications (surveillance), architecture, journalism and many other items found in the role. In initial research text-based image retrieval was performed, but with the advent of great challenges in text-based retrieval (eg spelling errors), content-based image retrieval has been introduced by researchers, which is by far the most effective method for image retrieval. Content-based image retrieval system uses features such as color, shape and texture. To extract the tissue properties local binary patterns and edge filtering methods are of particular popularity among researchers. A review of the methods presented so far shows that despite the quality of the descriptors and categories and retrieval methods, none of these methods can meet the needs and challenges of the present, so to improve the accuracy of image retrieval, in this study, a method introduced. To extract feature from the images, five color histogram descriptors, color moment, edge histogram, gradient oriented histogram and MRELBP were used. To classify the attributes extracted by the descriptors, three categories of support vector machine and k nearest neighbour and random forest are used. In the method, the features extracted by the five descriptors are combined and after classifying and identifying the test image class, using the Kmeans cluster, the closest images to the test image are retrieved from the identified class. Experimental results method on three databases Corel ۴k, Wang and Corel ۵k show We have accomplished the highest precision rate of ۸۶% using proposed CBIR system.

Keywords:

Image retrieval , Content-Based Image Retrieval methods , Local Image Features , Statistical features , Distance-based Identification