基于结构化字典学习的图像分类方法研究

发布时间：2018-05-12 00:16

本文选题：字典学习 + 图像分类　；参考：《天津大学》2016年硕士论文

【摘要】：图像分类是计算机视觉、多媒体、模式识别与机器学习领域非常活跃的研究方向。图像分类在很多领域有广泛应用,包括安防领域的人脸识别、智能视频分析、行人跟踪等,互联网领域的基于内容的图像检索、相册自动归类等。可以说,图像分类已经应用于人们日常生活的方方面面,计算机自动分类技术也在一定程度上减轻了人的负担,改变了人类的生活方式。近年来,字典学习被广泛应用于图像分类以及视频处理领域。在该模型中,图像被表达成一组字典的少量原子的线性组合。研究表明,在大多数图像处理应用中,例如人脸识别、对象识别、图像分类等,相较于非监督字典型学习方法,基于类别层次关系的监督型字典学习方法取得更好的实验性能。本文主要针对结构化字典学习模型及在图像分类方法中的应用进行了研究,主要研究内容如下:1.对非监督型字典学习和监督型字典学习进行概述,并分析现有字典学习理论在图像分类应用中存在的问题。2.引入一种基于图导向的监督型字典学习方法,该算法将类别层次结构的节点与节点的关系和边与边的关系嵌入到字典学习过程中,并将该算法应用到图像分类任务中,实验结果验证了该算法的有效性。3.基于图导向的字典学习方法存在的一些问题,引入一种基于边与层次关系的结构化字典学习方法,该算法通过建模将层与层之间的关系嵌入到学习过程中,实验结果表明该算法具有更好更稳定的性能。
[Abstract]:Image classification is an active research field in computer vision, multimedia, pattern recognition and machine learning. Image classification is widely used in many fields, including face recognition in security field, intelligent video analysis, pedestrian tracking, content-based image retrieval in the field of Internet, automatic classification of photo albums, and so on. It can be said that image classification has been applied to every aspect of people's daily life, and the computer automatic classification technology has lightened the burden of human beings to some extent and changed the way of life of human beings. In recent years, dictionary learning has been widely used in image classification and video processing. In this model, the image is expressed as a linear combination of a small number of atoms in a group of dictionaries. The research shows that in most image processing applications, such as face recognition, object recognition and image classification, the supervised dictionary learning method based on class hierarchical relationship has better experimental performance than the unsupervised word learning method. This paper mainly focuses on the structural dictionary learning model and its application in image classification. The main research contents are as follows: 1. This paper summarizes the unsupervised dictionary learning and supervised dictionary learning, and analyzes the problems existing in the application of the existing dictionary learning theory in image classification. This paper introduces a supervised dictionary learning method based on graph. The algorithm embed the relationship between node and the relationship between edge and edge of class hierarchy into the dictionary learning process, and applies the algorithm to the task of image classification. Experimental results show that the algorithm is effective. Based on some problems of graph-oriented dictionary learning method, a structured dictionary learning method based on edge-level relationship is introduced. The algorithm embeds the relationship between layers into the learning process by modeling. Experimental results show that the algorithm has better and more stable performance.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】