基于神经网络的多分区图像自动标注算法的研究与设计

发布时间：2018-04-24 07:44

本文选题：图像检索 + 图像自动标注　；参考：《内蒙古大学》2017年硕士论文

【摘要】：在文字搜索日益成熟的今天,图像搜索的需求也越来越大。但由于图像包含了更大的信息量,并且在存储方式上很难体现出图像的语义特征,所以图像的索引和检索显得十分困难。利用成熟的文字搜索方式,将图像搜索转换为文字搜索不失为是一个好的方法。而这种方式依赖于对图像语义的准确描述。这种描述方式随着图像数量的快速增长,手工实现无论从时间上还是费用上都太过昂贵,已经不能满足人们的需要。为了更好地实现图像检索,实现自动的生成标注图像信息的方法已经非常必要和紧迫。同时,图像自动标注算法在许多领域都有重要的应用。如针对视频的检索,针对盲人的电影剧情描述等等。本文使用神经网络的方法,对图像自动标注的方法进行了研究,具体内容如下:1、图像自动标注深深依赖于算法对图片对象识别的准确度。只有识别出图像中存在的对象,才能进一步的生成标注信息。而一张图像中的对象可能会有很多,每个对象有不同的属性,甚至有些对象会包含子对象,有些对象之间甚至可能会出现覆盖现象。本文方法首先要在众多的对象中进行筛选,选出重要的对象,才能进行识别。2、图像自动标注包含了对象识别模型和自然语言生成模型两个计算机视觉系统。对象识别模型使用了卷积神经网络模型进行处理。而自然语言模型则使用循环神经网络模型进行处理。两者共同组建成一个完整的系统。
[Abstract]:In the text search increasingly mature today, image search demand is also growing. However, because the image contains more information and it is difficult to reflect the semantic features of the image in the storage mode, it is very difficult to index and retrieve the image. It is a good method to transform image search into text search by using mature text search method. This approach depends on the accurate description of image semantics. With the rapid increase of the number of images, the manual implementation is too expensive in time and cost to meet the needs of people. In order to achieve better image retrieval, it is necessary and urgent to automatically generate and annotate image information. At the same time, automatic image tagging algorithm has important applications in many fields. Such as video retrieval, for blind film plot description and so on. In this paper, the neural network method is used to study the automatic image tagging method. The specific contents are as follows: 1. The automatic image tagging depends heavily on the accuracy of the algorithm for the recognition of image objects. Only when the objects in the image are identified can the tagging information be further generated. There may be many objects in an image, each object has different properties, even some objects may contain child objects, and some objects may even have overlay phenomenon. In this paper, first of all, we have to screen many objects and select the important objects before we can recognize .2. the automatic image tagging includes two computer vision systems: object recognition model and natural language generation model. The object recognition model is processed by convolution neural network model. The natural language model is processed by the circulatory neural network model. Together, they form a complete system.
【学位授予单位】：内蒙古大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP183

【参考文献】