中文摘要：

针对局部约束线性编码（LLC）没有考虑到特征描述符与视觉单词之间的显著相似关系问题,提出了结合显著相似关系的局部约束线性编码.通过显著kNN搜索方法和显著最大池方法将显著相似关系结合到LLC中,首先计算描述符和视觉单词之间的显著相似度,然后分别加入到kNN搜索方法和最大池方法中对LLC进行改进,最后在UIUC8等数据集上进行了实验.该方法相比传统LLC方法及其改进方法,图像分类精度有一定的提升.

英文摘要：

The important defect of locality-constrained linear coding（LLC）is not taking into account the saliency similarity relationship between feature descriptors and visual words.For this reason,locality-constrained linear coding combined with saliency similarity algorithm was proposed.The LLC was combined with saliency similarity relationship by two parts：saliency kNN（k-nearest neighbor）search method and saliency max pooling method.First,the saliency similarity relationship was computed between the descriptors and the visual words,and then which were added to the kNN search method and the max pooling method to improve LLC method.Finally,experimental was performed on UIUC8 and other data sets.Compared with traditional LLC method and its improvement work,image classification accuracy of the method has a certain improved.