为了解决大规模数据集下传统视觉词袋模型生成时间长、内存消耗大且分类精度低等问题,提出了基于监督核哈希(Supervised Hashing with Kernels,KSH)的视觉词袋模型.首先,提取图像的SIFT特征点,构造特征点样本集.然后,学习KSH函数,将距离相近的特征点映射成相同的哈希码,每一个哈希码代表聚类中心,构成视觉词典.最后,利用生成的视觉词典,将图像表示为直方图向量,并应用于图像分类.在标准数据集上的实验结果表明,该模型生成的视觉词典具有较好的区分度,有效地提高了图像分类的精度和效率.
Image Classification Based on Supervised Hashing with Kernels
In order to solve the problems of high generation time,large memory consumption and low classification accuracy of traditional bag of visual word(BoVW) model under large scale datasets,a BoVW model based on supervised hashing with kernels (KSH) is presented.Firstly,the feature point sample set was constructed by extracting the SIFT feature points of the image.Then learning KSH function,the similarity feature points were mapped to the same hash code,and each hash code represented the cluster center,so the visual dictionary was constituted.Finally,the generated visual dictionary was used to represent the image as a histogram vector and applied to image classification.The experimental results on the standard dataset show that the visual dictionary generated by the model has a good distinguishing degree,which effectively improves the accuracy and efficiency of image classification.
KSHbag of visual wordvisual dictionaryimage classification
国家科技期刊平台
NSTL
万方数据
维普
