Research And Implementation Of Image Automatic Annotation Based On Image Segmentation

With the rapid development of the Internet and the rise of mobile terminals,the number of images on the network is exponentially increasing,and the large-scale growth of image data provide more requirements for image understanding and processing.How to effectively realize the massive visual data organization and management,and how to accurately sort and retrieve these images have always been hot topics for scholars.Image retrieval has been widely used in scientific research and business.The users can quickly retrieve their own ideal pictures by reasonably classifying the image databases and establishing an effective retrieval mechanism.The existing mature image retrieval methods are mainly based on the text.The users search the images through the input semantic keywords.However,due to the existence of "semantic gap",the underlying characteristics of the image are often difficult to directly characterize the contents and the high-level semantic information of the image.In text-based image retrieval,if the keyword can accurately express the semantic content of the image,the semantic gap can be reduced to the maximum extent,so that the target image can be retrieved accurately.The image automatic annotation technology can be used to understand the high-level semantic information from the image contents,obtain the image information keywords,and automatically mark the image,so it becomes the key of image retrieval.The image automatic annotation technologies are studied in this paper.In order to improve the precision of annotation,The main work of this paper is as follows:(1)An improved image segmentation algorithm is proposed in the paper.This method uses the flow procedure of JSEG image segmentation method,such as filter,color quantification,access to seed area,regional growth,regional merging and so on.However,the segmentation of JSEG image only considers the color feature of the image,and it can't completely express the characteristics of the image.To address this problem,the image segmentation method based on color and texture is proposed in this paper.The extraction of texture features makes it possible to segment the image more accurately,thus improves the effect of image annotation.(2)An automatic image annotation system is designed and implemented in this paper.By segmenting the image,the underlying characteristic information of each feature region is obtained(each region is represented by an eigenvector).Then establish the underlying characteristic information and the mapping relationship between the underlying characteristic information and the high-level semantics.The support vector machine(SVM)is selected to train a model,and then classify the test images using the training model.By querying the mapping relationship,it can obtain the high-level semantic information of the images and realize the automatic annotation of the images.