【摘要】 利用基函数稀疏性的极大值特点,本文提出一种改进的基于独立分量分析的图像特征提取算法。通过分析图像的拉普拉斯先验条件,独立成分分析的问题简化为L1范数求最小的问题,但是采用其对偶空间L∞范数求极大值的方法解决问题更加容易。相对其它独立成分分析方法,本文方法不需要对高阶非线性对比函数估值进行复杂的优化。实验结果表明,本文算法相比其它独立成分分析算法,具有更好的稀疏性和更快的收敛速度。更多还原

【Abstract】 In this paper, an improved algorithm of image feature extracting for independent component analysis was proposed based on basic functions’ maximization of sparseness. Starting from a Laplacian Priori of the image, the ICA problem was boiled down to a minimum of L1 norm problem, but the problem would be much easier to solve by searching a maximum of its dual space L∞ norm. This method avoids the expensive optimization of high-order non-linear contrast function, which can be commonly found in other ICA methods. The simulation results show the proposed method has sparser and faster convergence than others.更多还原