基于PCA-SIFT特征与贝叶斯决策的图像分类算法

摘要

In order to cope with the problems that existing SIFT-based image classification algorithms require a large amount of storage space and are sensitive to image backgrounds,this paper presents a novel image classification algorithm which is based on PCA-SIFT features and Bayesian decision.The algorithm first applies the principal component analysis (PCA)to reduce the dimensionality of SIFT from 128 to 36,in training process,it makes regional matching on PCA-SIFT descriptors of the training sample images.In order to improve its robustness on background image interference,we selected the stable PCA-SIFT descriptors in object images based on their matching rates,and then used the maximum likelihood estimation to estimate the probability distribution parameters.Finally we used Bayesian decision theory to implement the image classification.Simulation experiment showed that this algorithm has higher classification accuracy compared with existing SIFT-based image classification methods.It also has minimum storage space requirement and higher computation efficiency.%针对现有的基于SIFT特征的图像分类算法具有较大的储存空间需求及对图像背景较为敏感的问题，提出一种基于PCA-SIFT特征和贝叶斯决策的图像分类算法。该算法首先应用主成分分析将SIFT特征从128维降为36维，在训练过程中，对训练样本图像的PCA-SIFT进行区域匹配。基于匹配率选择目标图像中的稳定PCA-SIFT以提高算法对背景图像干扰的鲁棒性，然后应用最大似然估计估计概率分布参数，最后，应用贝叶斯决策理论实现图像分类。仿真实验表明，该算法与现有的SIFT图像分类算法相比分类精度高，而且具有最小的储存空间需求与较高的计算效率。