基于非负稀疏编码的图像特征提取及应用

Image Feature Extraction and Using Based on Non-negative Sparse Coding

NNSC是把多维数据分解成非负稀疏分量的一种方法,而且这种方法已成功地应用于建模视觉感知系统的感受野。通过仿真实验证明了NNSC方法在自然图像的特征提取中的有效性;而且与独立分量分析(ICA)方法相比,实验结果证明NNSC法提取的特征基要优于ICA基,其图像重构目视效果明显好于用ICA基恢复的结果。

The paper mainly discusses the application of Non-negative Sparse Coding （NNSC） in natural image feature extraction. NNSC is a method for decomposing multi-variants data into non-negative sparse components; more over, this method had been used successfully to model receptive fields of vision-perceptional system. The paper refers to this approach in image feature extraction, and the reconstruction image is obtained by those learned feature basis vectors and non-negative sparse components. Finally, Simulations demonstrate the effectiveness of natural image feature extraction using NNSC. More over, comparison with ICA, test results prove that feature basis vectors extracted by NNSC outperform those extracted by ICA, and the visual effect of the reconstruction image by NNSC is distinctly better than that by ICA.