Research On Image Retrieval Based On SVM Semantic Classification And Visual Features Extraction

Together with the fast development of multimedia and computer network technology, image database is also expanding in an astoundingly large scale. How to find the image you need in the ocean of image database is now a crucial problem which is frequently researched among many other subjects in computer science. As there is huge gap between high-level visual features of an image and its in-built high-level semantics, image retrieval based on either of the above mentioned features is often ineffective.In this paper, the author will venture to bring forward a new image retrieval method which combines both semantic feature retrieval method and visual features retrieval method, used the image database classification to filter the retrieval results. There is also careful study on visual features extraction, semantics retrieval and image classification based on learning mechanism.The first part gives a brief analysis on three kinds of image retrieval methods currently used in research, especially on supervised and non-supervised methods on semantic feature retrieval. Semantic meaning is important in conveying the content of an image, and image classification based on semantic meaning is helpful to improve the performance of image retrieval.The following part focuses on visual features extraction and many elements used in that process including color, texture and shape. As images on natural scenery are often rich and complex in content, the author adopted image features based on HSV(Hue, Saturation, Value)color, texture of co-occurrence matrix representation and gray texture moments and shape of moment invariants. All the above mentioned retrieval methods are proved in experiment.The last part mainly studies the classification method of support vector machine, with stresses on the selection of kernel functions and there parameters, and further develops an image classification model, based on combined study of visual features and semantics, with SVM classification method, to retrieve image from image database. Experiment conducted with image database on nature scenery has proved the combined image retrieval method, put forward by the author, is better off in terms of effectiveness and accuracy.