Research On Context - Based Image Comprehension Algorithm

With the development of modern society and the improvement of people’s living standard,computer technology has also developed rapidly, so the study of image is becoming more andmore deeply. Images has been an important part in our daily life. We can see imageseverywhere in our daily life, such as the photos taken in our modern life or by video monitoringsystem, the images taken by satellite and so on. So, image has an important influence on ourdaily life. Computer technology has developed rapidly in recent decades. Combining imageprocessing with computer technology can make the process simple. This provides a lot offavorable condition for image processing.Image understanding is to label the corresponding object in the image and record therelative position of the object in the image. Finally, it outputs the corresponding target and areawith tags. It is a process that the computer understand the image by itself, just like the human.People obtain the images by visual, then they analysis the image with brain, thus to get theinformation of the image. Computer can get images in several ways, such as the camera, themonitor, the satellite, the infrared scanner and so on. Image understanding is to process theobtained images with the image processing method and output the tagged information. With thedevelopment of computer vision and artificial intelligence, some new researches constantlyemerging. Image understanding is widely used at more and more areas, such as industrial vision,human-computer interaction, biological visual field and so on.Because of that the images that we got from our daily life always have some noise, sobefore introducing the main content of this article, this paper put forward an image pretreatmentprocess, that is an image enhancement algorithm. This algorithm combined the traditionalRetinex algorithm and the color statistical. This image enhancement algorithm can make therecognition process and the image understanding process more accurate.Then, we introduced a scene recognition algorithm, because of that scene recognition is animportant part of image understanding. On the basis of the previous algorithm, we proposed anew scene recognition algorithm. The steps is patch match, action recognition and gazeprediction. The experiment can demonstrate the effectiveness of this algorithm.Next, this paper gives the algorithm of image understanding that based on context. Thispaper segmented the image at first and build the PatchNet for the image. Then it recognizes theactions of the people in the image and gives the tag transfer. Finally, it restores the depthinformation of the image. So, knowing the actions of the characters, the context of the image and the depth information of the image, we can give the understanding result.Finally，this paper summarizes the research content of this paper and briefly introduces theadvantages and the disadvantages of the algorithms of this paper. Then, it present the work ofthis paper for some time to come.