基于形状特征的目标检测算法研究

发布时间：2018-12-18 14:14

【摘要】：手机照相技术的发展和数码相机价格的降低，使越来越多的消费者拥有自己的拍照手机和数码相机，并用它们记录生活中的点点滴滴。图片分享网站的兴起，让人们能够将照片上传到这些网站上，与家人和朋友们分享。另外，人们可以利用各大搜索引擎对网络上公开的图片进行搜索。现有的图片搜索技术大多都是基于图像附带的文本信息或者颜色统计之类的简单特征。利用计算机视觉技术对图片的内容进行精确理解，是图片搜索的发展趋势。目标检测是计算机视觉领域的一个重要研究方向。其主要任务是定位并识别图像中所包含的目标，从而为场景识别提供信息。目标检测技术虽然已经取得极大的发展，但离实际应用仍有一段距离，仍然是一个很值得研究的方向。观察角度的不同，以及同类别的不同目标的外观存在着很大的差异，给目标检测带来了极大的难度。目标检测算法需要极大的计算需求，这也限制了目标检测的实际应用。现有的目标检测主要是基于局部纹理特征。利用形状特征分析图像，，可以减少图像特征的冗余性。而对基于形状特征的目标检测的研究仍然较少。本文对基于形状特征的目标检测算法进行研究，并构建了一个完整的目标检测系统。针对基于形状特征高度依赖于边缘方向的特性，该系统选用了能够很好保存边缘信息的Amit边缘检测算法来提取图像边缘信息。该系统还使用基于形状的检测子和多种描述子来描述局部形状特征，并用最新的特征包方法来整合局部形状特征。最后，本文在Cartoon30数据集和VOC2012数据集上对所提出的目标检测算法进行测试和评价。
[Abstract]:With the development of mobile camera technology and the decrease of the price of digital camera, more and more consumers have their own photo phone and digital camera, and use them to record the bits and pieces of their lives. The rise of photo-sharing sites allows people to upload photos to these sites and share them with family and friends. In addition, people can use each major search engine to search for pictures published on the network. Most of the existing image search techniques are based on simple features such as text information or color statistics. It is the development trend of image search to use computer vision technology to accurately understand the content of pictures. Target detection is an important research direction in the field of computer vision. Its main task is to locate and recognize the targets contained in the image, so as to provide information for scene recognition. Although the technology of target detection has made great progress, it is still a long way from the practical application, and it is still a research direction. There are great differences in the appearance of different targets from different points of view and the same category, which brings great difficulty to target detection. Target detection algorithm needs a lot of computing requirements, which limits the practical application of target detection. The existing target detection is mainly based on local texture features. The redundancy of image features can be reduced by using shape features to analyze images. However, the research on shape feature based target detection is still few. In this paper, the object detection algorithm based on shape feature is studied, and a complete target detection system is constructed. Aiming at the feature that shape feature is highly dependent on edge direction, the system chooses Amit edge detection algorithm which can save edge information well to extract image edge information. The system also uses shape based detectors and multiple descriptors to describe local shape features and integrates local shape features with the latest feature packet method. Finally, the proposed target detection algorithm is tested and evaluated on Cartoon30 data sets and VOC2012 datasets.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.41

【共引文献】