Research On Content-based Image Retrieval Technology

In recent years, because of the rapid development of multimedia, the Internet technology as well as the widely used social apps, the image with it rich information is becoming more and more important in our daily life. Also at the same time the huge number and the rapid growing rate of the images have never seen before. As we are in an era of which efficiency is the first priority. It is not only practical significance but also useful to extract the information we need fast and preciseness.Image can express information more intutive and itself do have much more meaning than words. Image occupied a very essential positon in ways people obtain information. But the technic of image recognition in computer is still far from the ability of man brain which can directly understand the meaning of the image. Semantic-based image retrieval is the ideal target. The common method to search the image is text-based image retrieval (TBIR) which use the text around the image to help understand the real meaning of the image; the content-based image retrieval (CBIR) which use the physical visual features such as color, shape to compare and search images. Because of image contains a wealth of features like color, shape, so that the image is more complex and time-consuming in storage and calculation compared to text data. We considered both the advantages and disadvantages of the IBIR and CBIR method, and propose a fast content-based image retrieval method in order to retrieve images quickly and accurately by make full use of the text meaning, the image visual features and the parallel computing cloud platform. The main content and innovation of this paper is as follow:1. First, extract the keywords from the text around the images, and then build the similar image collection of the target image based on the extracted keywords. We improved the TextRank keywords extraction algorithm here, we add the sematic meaning of the word while the word is calculation its weight, by this way the extracted keywords contains both semantic meaning and probability in its weight. Then the keyword can express the image better, remove the irrelevant image and narrow my search for the parallel image matching in next step.2. Because we use the same procedure to match images, so we can change the repeat process into parallel calculation. And there will be more and more images need to be retrieved in cloud parallel platform. So after the first step, we use the most popular MapReduce framework on Hadoop platform to do the parallel calculate work. In order to make the image retrieval process more efficient.