基于聚类的智能图像分析算法毕业设计论文北京邮电大学.docVIP

基于聚类的智能图像分析算法 摘　要 智能图像处理技术在众多领域得到了广泛应用。在具有监控、报警等功能的安防系统中，在自然基因显微系统中，在模式识别系统中等，智能图像处理技术都起到了不可小觑的作用。 目前基于内容的智能图像识别与分类技术在准确性具体应用方面还面临着许多难题。本文通过介绍智能图像分析方法及相关算法理论，重点研究以SVM算法为代表的监督分类算法及以k-means聚类算法为代表的费监督分类算法，并结合Hu图像矩不变特征，对图像进行聚类分析及分类。在理论学习的基础上，运用MatLab实现算法并验证应用效果。 有监督分类方面，本文采用了提取能够较好的保持图像的边缘、形状等特性的Hu矩不变特征作为训练特征，分类方法采用了基于聚类的SVM算法。在提取出训练样本的特征值后，将其输入SVM的训练网络进行训练。最后将待分类图片输入即可得到分类结果。本文计算出了该非监督分类方式分类结果的准确性，并对其进行了分析与讨论。 无监督分类方面，本文采用了k-means分类方法。预先设定好分类的类别数后，输入待分类图片，则系统通过调用分类函数，将自动分类的结果输出。 在算法研究的基础上，设计并实现了水果图像智能分析应用系统，具有创建特征值数据库、创建训练网络、图像有监督分类和图像无监督分类等功能。当进行图像有监督分类，即SVM算发分类时，准确率可达到将近70%。 关键词 SVM k-means　图像分类 Intelligent?Image Analysis?Based on Clustering?Algorithm ABSTRACT Intelligent image processing technology has been widely applied in many fields. In monitoring and alarm security system, in natural gene microscope, and in the middle pattern recognition system, intelligent image processing technology has played highly important role. Currently content-based image recognition and classification of intelligent technology are facing many problems in specific application for accuracy. This paper will describes intelligent image analysis method and algorithm theory, meanwhile combines with the same characteristics of HU image moments, and focuses on the SVM algorithm for classification and supervision of representatives of the costs of supervised classification algorithms. In the theoretical study, verify the application of results based on the use of MatLab algorithm. In the phase of supervising classification, this paper used Hu moments invariant feature as a training feature that can keep the extracted image edge, shape and other characteristics using SVM-based clustering algorithm. After extracting samples’ characteristic value, put into SVM’s training network to have training. Finally the input image can be classified by the classification results. This paper concludes the approach to the classification of non-supervised classification accuracy of the results meanwhile an