As a composite sintered material, polycrystalline diamond compact (PDC) is made from polycrystalline diamond layer (PCD) and tungsten carbide (WC) alloy at high pressure and high temperature (HPHT) conditions. Defects such as crack, white spots and white edge at the surface of PCD are unavoidable in the manufacturing process, which influences the appearance and performance of the product. An automatic and non-destructive method was proposed for the accurate identification and classification of surface defects. The method is based on machine vision technique and support vector machine (SVM). First, the defect models were established. In order to obtain the region of polycrystalline diamond layer, a local boundary extraction method in terms of the histogram projection gradient extremes was utilized. Then, the accurate detection of defects were realized by image filtering and feature extraction. Accordingly, seven defect features were selected as input vectors of SVM. Finally, 450 samples were trained and tested, and polynomial kernel function was selected as the kernel function of SVM model. The results show that the SVM model with optimal parameters provided a classification accuracy of 99%. The experimental results indicate that the proposed method provides defect quantification with reasonable accuracy facing various surface defects, and it provides an effective defect detection method for PDC instead of manual way that can be used for automatic detection.