基于ICA的特征不变性目标识别

在图像目标识别中，目标图像平移、尺度和旋转不变性是一个重要前提，而高阶矩特征存在稳定性差的缺点．提出采用主分量分析(PCA)方法确定目标在最大方差意义下主轴的旋转角度，并结合稳定性好的低阶矩特征实现目标的平移、尺度和旋转不变性变换；然后利用独立分量分析(ICA)良好的目标特征抽取能力，在各目标特征空间重建目标模型，并通过对重建模型的误差分析实现目标识别；最后通过PCA确定目标旋转角度测试和ICA目标识别测试两个实验，证实了本文算法的鲁棒性和准确性．

Invariance-Based Target Recognition Using Independent Component Analysis

In target recognition, translation-, scaling-and rotation-invariance is a very important factor. While there exist object moment invariance, the high-order ones are sensitive to noise, resulting in instability in target recognition. To overcome these difficulties, a novel approach is proposed in this paper. First, principal component analysis (PCA) is used to determine a target's major axis whose direction has the maximal variance. Combined with stable low-order moments, the translation, scaling and rotation invariance of a target is achieved. Independent component analysis (ICA) is used to extract target features for different classes of targets. Next, target models are reconstructed in their own feature spaces. Finally, target recognition is conducted based on reconstructed models' error analysis. The proposed algorithm is tested with two experiments. The experimental results demonstrate the robustness and accuracy of the proposed algorithm.