基于手绘草图的图像检索技术与算法实现

发布时间：2018-11-11 21:09

【摘要】：随着数字图像库中图像的扩增和触屏手机、平板电脑等高科技硬件的普及,用以图找图方式进行图像检索的技术逐渐进入到人们的生活中。近些年,越来越多的国内外学者涌入到手绘草图检索系统的研究中,并且取得了一定的成果。从完全依靠文字进行检索,到基于图像的关键字检索,再到完全依靠手绘草图的检索,图像检索技术经历了跨越式的发展。在草图检索系统中,人们可以在系统提供的画板上对目的图像进行简单的勾勒,然后系统会对草图进行边缘提取、轮廓跟踪、特征点分配、特征转换、相似度匹配等处理,最终为用户提供出几幅相似度较高的目的图像。在搜索目的图像的过程中,用户搜索的主要依据是图像的形状信息,这也符合人眼观察的特点。因此在草图检索系统中,往往只考虑图像的轮廓特征,而不考虑图像的颜色特征和纹理特征。本文在研究了国内外先进草图检索技术的基础上,对草图检索的相关算法进行了改进,设计出了一个新型的图像检索系统。首先,针对用户输入草图的特点,增加了草图预处理模块;其次,本文对比了大量的边缘提取算法,并且创新性的对王福生等人提出的五邻域跟踪法进行了改进,提出了x轴和y轴方向三邻域跟踪法。改进的三邻域跟踪法不仅降低了系统的时间消耗,还提高了轮廓点跟踪的正确率;然后,本文将形状上下文算法和傅里叶变换相结合,使改进的形状上下文算法具有了旋转不变性,解决了图像偏移会降低检索正确率的问题;最后,本文按照改进的算法,设计出了一个小型的手绘草图检索系统,并用一个具有300幅图像的小型图库进行了实验,取得了良好的效果。
[Abstract]:With the expansion of image in digital image library and the popularization of high-tech hardware such as touch-screen mobile phone and tablet computer, the technology of image retrieval in graphic search has gradually entered into people's life. In recent years, more and more scholars at home and abroad pour into the research of hand-drawn sketch retrieval system, and have achieved certain results. Image retrieval technology has experienced a leap-forward development from text retrieval to image-based keyword retrieval and to hand-drawn sketch retrieval. In the sketch retrieval system, people can simply outline the target image on the drawing board provided by the system, and then the system will deal with the sketch edge extraction, contour tracking, feature point assignment, feature conversion, similarity matching and so on. Finally, several target images with high similarity are provided for users. In the process of searching the target image, the main basis of user search is the shape information of the image, which is also in line with the characteristics of human eye observation. Therefore, in the sketch retrieval system, only the contour features of the image are considered, but the color and texture features of the image are not considered. On the basis of studying the advanced sketch retrieval technology at home and abroad, this paper improves the relevant algorithms of sketch retrieval, and designs a new image retrieval system. Firstly, according to the characteristics of user input sketch, the sketch preprocessing module is added. Secondly, this paper compares a large number of edge detection algorithms, and innovatively improves the five-neighborhood tracking method proposed by Wang Fusheng and others, and proposes a three-neighborhood tracking method in the direction of x axis and y axis. The improved three-neighborhood tracking method not only reduces the time consumption of the system, but also improves the accuracy of contour tracking. Then, the shape context algorithm and Fourier transform are combined to make the improved shape context algorithm rotation-invariant, which solves the problem that image offset will reduce the retrieval accuracy. Finally, according to the improved algorithm, a small hand-drawn sketch retrieval system is designed, and a small image library with 300 images is used to carry out experiments, and good results are obtained.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】