基于异质局部特征的图像稀疏表示

为了有效描述图像的多角度视觉内容,提出一种将图像异质局部特征集通过稀疏学习映射为图像全局稀疏表示的新方法.该方法从不同的训练特征集中学习超完备视觉词典,经过局部稀疏编码、最大值合并、加权联接及归一化等一系列处理步骤融合多种局部特征的互补信息,最终形成一个高维稀疏向量来描述图像的多角度视觉内容.将其应用于基于内容的图像检索(CBIR)任务中,实验结果表明,这种基于异质局部特征学习而来的图像全局稀疏表示解决了单一局部特征集描述图像的局限性和高维局部特征集相似性度量时空复杂度高的问题.

Image Sparse Representation Based on Heterogeneous Local Features

To effectively describe the multiview visual content of images, a new method of mapping a set of heterogeneous local features into a holistic image sparse representation via sparse learning is presented in this paper. It learns overcomplete visual dictionaries from different training feature sets and integrates complementary information from multiple image local features to produce a final high-dimensional sparse vector through a series of processing steps such as local sparse coding, max pooling, weighted concatenating and normalization. This method is used for the task of Content-Based Image Retrieval (CBIR) and the experimental results indicate that this holistic sparse representation learnt from heterogeneous local features can solve the problems of the limitations caused by a set of single local features for describing image and the high time and space complexity caused by the similarity measurement between sets of high-dimensional local features.