Research On Efficient Region Segmentation Method Of Image Annotation

With the advent of4G era, The tide of mobile Internet has quietly come. People’s work and life have been accompanied by intelligent terminals all the time. The most important and the most common intelligent terminal is the appearance of the smart phone, which has been becoming an integral part of people’s daily lives. Along with the hot of APP, such as MicroBlog, WeChat and the advances in storage technology, taking pictures conveniently and sharing them in the circle of friends have become the most fashionable thing. It meets people’s spiritual enjoyment, but it also causes the geometric growth of the number of pictures. And in our computers, cell phones and the Internet there are a lot of digital pictures. So how to find a picture which we really want is an urgent problem. This leads to the semantic tagging problem which will be studied in this dissertation. Only if the image is marked by the words which truly reflect the picture content, coincide with the human vision and easily be understood, so it can be quickly retrieved by searching techniques. Obviously using artificial method to mark each image is unrealistic. Only if the correlation between the image content and the words is founded the computer can automatically complete the marking work. There are a lot of automatic image annotation methods and they are mainly divided into two categories: One is to allow users to participate in image annotation and retrieval; Another way is to identify the possible correspondence between underlying features and high level semantics on the smaller image size. Obviously the second category using computer to finish the semantic annotation and retrieval is the future trends. The core problem with this approach is to find the optimal image segmentation algorithm. After the segmentation, it gets more accurate segmentation region and it is easier to get the feature information. So it can obtain good semantic annotation results.The main contributions of this paper are as follows:1. The process in which the calculation of J value and the selecting of seeds and regional growth repeatedly in JSEG image segmentation algorithm has been improved. This reduces the complexity of the algorithm and makes it easy to be used in the late image semantic annotation system. Therefore, it makes the image processing fast and efficient, and also makes the algorithm more practical.2. The post region merger method in JSEG algorithm has been improved. The original JSEG algorithm is simple to take the color histogram for combining, without considering the correlation between the respective areas. But this paper introduces the idea of graph theory. The method is to make the divided region correspond to the graph nodes, and operate the secondary image segmentation by graph theory ideas.3. Proposed an efficient regional segmentation algorithm which is suitable for semantic annotation. After dividing, it is easy for feature extraction and image annotation using classification methods. Experimental results show that the accuracy of image semantic annotation system which uses the new segmentation algorithm is relatively high.