稀疏表示的图像分类研究综述

良好的特征提取方法能减轻后续图像分类与识别的工作量。针对具体的分类问题提出了不同的特征提取方法,并在图像分类和识别任务上取得了较好的效果。然而,已有的基于传统方法的特征提取存在一些明显不足,即随着视觉任务规模的增大,直接利用这些传统方法进行特征分类,效果并不理想。提出的特征表达方法,在图像最基本特征基础上进行矢量量化、稀疏编码或其它表达以形成一幅图像最后的特征。着重介绍基于稀疏表示的特征分类算法并对其进行分析,最后探讨存在的问题和今后研究的方向。

Survey of Image Classification Based on Sparse Representation

Good feature extraction method can reduce the workload of subsequent image classification and recognition. Different feature extraction methods are proposed for the specific classification problem, and achieved good results in image classification and recognition tasks. However, there are some obvious shortcomings of the existing feature extraction based on the traditional method. With the increasing of the size of the visual task, direct use of these traditional methods for feature classification is not ideal. The feature expression method is proposed, which is based on the most basic features of the image, and the sparse encoding or other expressions are proposed to form a final image. Based on sparse representation and its analysis, this paper focused on the feature classification algorithm and finally discussed the existing problems and future research directions.