A Novel Approach to Edge Detection Based on PCA

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-Loeve 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.