一种基于约束线性编码的图像分类改进算法  

An Improved Algorithm of Image Classification Based on Constrained Linear Coding

作　　者：胡广平[1] 周华强[2] 

机构地区：[1]河南工程学院,郑州451191 [2]中原工学院,郑州450007

出　　处：《科学技术与工程》2013年第28期8329-8332,共4页Science Technology and Engineering

基　　金：国家自然科学基金项目(71173248)资助

摘　　要：针对图像分类中量化编码的问题,提出了一种简单而且高效编码方法,叫做局部约束线性编码(Locality-Constrained Linear Coding)算法;并将其应用在传统空间金字塔模型(Spatial Pyramid Matching)的向量量化(Vector Quantization)中。通过使用局部约束,LLC先将图像描述子映射到它的局部坐标系统,然后再将这些映射后的坐标经过Max-pooling整合成最终的图像表示。传统的基于词袋模型(Bag-of-World)的空间金字塔需要使用非线性分类器才能获得较好的分类效果,而LLC通过使用线性分类器获得了比传统非线性SPM更好的分类效果。For the problem of quantization coding in image classification, a simple and efficient encoding meth- od called Local-constrained Linear Coding is proposed, which makes it apply in the Vector Quantization in tradi- tional Spatial Pyramid Model. Using local constraints, LLC project each descriptor into its local-coordinate system, and the projected coordinates are integrated by max-pooling to produce the final representation. With the linear classifier, the proposed approach performs outstandingly better than the traditional nonlinear SPM. In addition, a fast approximation LLC algorithm be also proposed to reduce the amount of computation in LLC.

关 键 词：图像分类 局部约束线性编码 向量量化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]