面向盲人的图片自动描述系统的研究与实现-计算机应用技术专业毕业论文.docxVIP

浙江大学创，1 t:学位论文 Abstract Abstract In the cuηent portal page，a large number of images have been inserted into the text to facilitate quick reading and deepen the readers understanding of the text.丁here are a large part of these images without any labeling，which gives blind people access to information has brought great difficulties. The traditional method of image auto-annotation can produce a number of content-related words，in general terms sighted enough，but for blind users still cant get enough of the picture information. However，the context at the picture contains many texts and graphics-related information ，so using the automatically marked words， the image can be described in a similar way to automatically extract the part of the original sentence. As a description of the image，no doubt greatly increase the picture information for the blind access. Range of network types and quantities of images，a fixed number of standard image set cant meet the mark set of image sample database as the network requirements. The social networking site Flickr can be usedωa sample database to meet the requirements of the source image，the image tags can also be seen as a mark of different users of the word image，the label recommends that image automatic annotation process. While improving the accuracy of label ranking is based on recommendations generated labels. Based on these observations ，we propose a method described for images in the text illustrations automatically. Start with the Flickr Web site to crawl image as a sample set，and sort the image tag，and then using the recommended tags to the image extraction process in the context of abstracts，and finally output as a description of the lmage. Algorithm analysis and experimental results show that the method used for tags sorting and recommendation is effective. Description of the final extracted image can be more accurately express the image