为了更有效地进行边缘检测，通过分析PCA的方向特性，提出了一种基于PCA的边缘检测方法。PCA先利用KL变换来将原始数据变换成维数较少的特征数据，该变换在能量积聚和数据取舍上都具有方向性;同时在证明PCA的这两个方向特性的基础上，提出了一个经两次PCA操作获取边缘的新方法——TPCA。该新方法首先通过对图像进行PCA来得到其重建后的残差；然后再对该图像的转置图像进行PCA，并将所得残差做转置；最后通过对两个残差进行叠加,并二值化来得到比较好的边缘。实验结果表明，该算法不仅有效稳定，而且与经典的边缘检测算子相比，在提取感兴趣区域方面有独特的优势。

We present a PCA-based edge detection method with analysis on the orientation character of PCA. PCA translates the original data set to feature components in low dimension space using Karhunen-Loéve transform，which shows the tendency on energy collection and data selection. We point out and prove these orientation characters，and then present the new detection method TPCA，which processes an image with twice principal component analysis. First，an image is analyzed with PCA，and the residual is retained. Then，the image’s transpose is processed using PCA again，and the residual is transposed too. Finally，the two residuals are added. A better edge will be producted just with some simple operates，such as binary process. Experimental results show that the algorithm is effective,stable and has its own advantages compared with the traditional algorithms.