基于二进制哈希与空间金字塔的视觉词袋模型生成方法

摘要

构建视觉词典是视觉词袋模型中的关键步骤,目前大多数视觉词典是基于k-means及其改进算法聚类生成.但由于k-means聚类的局限性以及样本空间结构的复杂性与高维性,该方式构建的视觉词典存在区分性较差、构建时间过长、不包含空间信息等问题.为此,提出一种改进的视觉词袋模型生成方法,以缩短视觉词典的构建时间.提取图像的局部特征点,构成局部特征点描述集.学习二进制哈希函数,将局部特征点映射为视觉单词,并对视觉词进行过滤,生成二进制哈希码的视觉词典.利用生成的视觉词典,结合空间金字塔匹配模型生成新的视觉词典模型,将图像表示为空间金字塔直方图向量,并应用于图像分类和检索.实验结果表明,该模型具有较高的分类精度和检索性能.%Constructing visual vocabulary in the Bag of Visual Word(BoVW)model is a critical step,most visual vocabulary is generated by the k-means algorithm or its improved algorithm.Because of the limitation of the k-means algorithm and the complexity and the high-dimensionality of the sample space,visual vocabulary generated by these methods have the problem of low discriminative long running time and without space information.For these problems,a BoVW model is proposed based on binary Hashing and space pyramid,which can shorten the visual vocabulary generation time sharply.It extracts the local feature points from the images,learns binary Hashing functions,which map the local feature points into visual words,filters the visual words and generates the visual vocabulary whose visual word is binary hash code.The new BoVW model is composed with the visual vocabulary and Spatial Pyramid Matching(SPM)model,which represents the images by the histogram vector of space pyramid and is applied in image classification and retrieval.Experimental results on the common datasets show that the model has higher classification accuracy and retrieval performance.