A completely unsupervised defect detection method is proposed based on the Gabor filters and Mean Shift clustering to achieve the accurate automatic defect detection of a FPC gold surface. Firstly

the multi-dimension characteristics of an image to be detected are extracted by a series of processing steps including Gabor filter banks

morphological open and Gaussian smoothing. Then

the Principal Component Analysis (PCA) is used to reduce the pixel characteristics from multi-dimension to 2-D for reducing computation time in the next clustering. Finally

Mean Shift method is applied to cluster pixels with 2-D characteristics and the results can be divided into defect and non-defect groups to produce the binary image. The whole process needs to neither predefine the type of defects nor the texture type of FPC gold surface

which can be defined as a completely unsupervised method of detecting defects. A number of images of FPC gold surfaces with a variety of defects have been tested. Detection results show that the proposed method can accurately separate all types of defect regions from the background and has the characteristics of high stability and strong ability to identify defects.