To solve the problem of high missing rate and complicated operation in the detection methods of faulty insulator, this paper proposed a new method to automatically extract the disks and steel caps area of insulator from infrared image: via infrared image preprocessing, feature extraction, angle correction and regional extraction to achieve the purpose. This method first makes gray scene processing of insulator infrared image, noise reduction and binarization; then extracts the feature set related to insulator from the binary image to correct the angle of the binary image; finally extracts the disks and steel caps area through the specific algorithm of regional extraction. By extracting the absolute temperature region, texture and relative temperature difference as the insulator state recognition feature set, and the information obtained by measuring the voltage distribution of insulator state as output vector, the model was optimized by training. And the 220kV experiment confirms the validity and usefulness of the method.