End Face Recognition and Positioning Method in Unstable Definition Image based on Cross Entropy Estimation

Abstract

The current image recognition algorithm uses the preset operator to extract the edge and uses the Hough Transform to recognize the end face, which is not applicable to the case where the image definition is unstable. This paper presents a method for recognizing and positioning the end faces in unstable definition images based on cross entropy estimation. Gaussian Mixture Model is established for the image intensity and the end face image is segmented based on the model parameters, which are estimated by EM algorithm. A probability distribution is generated to characterize the intensity distribution of the image near the end face. By calculating and minimizing the cross entropy of the image intensity distribution and the designed probability distribution, the position of the end face image feature is estimated. The experimental results show that the proposed method can extract all circular features from 30 test images within 5mm range of axial position error, which shows high robustness and adaptability. The estimated position errors of the centers of end faces are within 1/3 radius, which meets the requirements of automatic assembly process.