摘要
为提高局部约束线性编码(locality-constrained linear coding,LLC)的效率,提出一种结合邻居匹配策略改进的LLC方法。依据输入向量的空间相关性,在采用LLC方法计算输入向量的近邻码值矩阵之前,计算输入向量与空间相邻的已编码输入向量之间的欧氏距离,用其推断输入向量与码本中所有码值之间欧氏距离的上下边界,依据距离下边界判决条件跳过部分码值与输入向量的距离计算,依据距离上边界快速求解输入向量的近似近邻码值矩阵,依据LLC方法进行向量编码。图像分类实验结果表明,该方法的分类正确率高,编码耗时少。
To improve the efficiency of locality-constrained linear coding （LLC） method, a modified LLC method improved by neighborhood matching was proposed. According to the spatial correlation of input vectors, the Euclidean distance was computed between the current input vector and its neighborhood input vectors that had encoded, before computing the nearest codeword matrix of the input vector using LLC method, to infer the up and down bounds of Euclidean distances between the current input vector and all codewords in codebook. Some distance computation between current input vector and part of codewords was skipped according to the down bound of distances, and the approximate nearest codeword matrix of the input vector was fast computed according to the up bound of distances, and LLC method was used to encode the input vector. Experimental results on image classification show that, the proposed method has high classification accuracy and less time-consumption of coding.
作者
田广强
张岐山
TIAN Guang-qiang ZHANG Qi-shan(College of Mechanical and Electronic Engineering, Huanghe Jiaotong University, Jiaozuo 454950, Chin)
出处
《计算机工程与设计》
北大核心
2017年第8期2217-2221,2261,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(034031122)
河南省科技攻关重点计划基金项目(122102210563
132102210215)
河南省高等学校重点科研项目计划基金项目(15B520008)
关键词
图像分类
局部约束线性编码
向量量化
码本
邻居匹配
image classification
locality-constrained linear coding
vector quantization
codebook
neighborhood matching
