Automatic segmentation of FLIR image based on gradient operator

In the FLIR automatic infrared object tracing system, the gradient histogram of the FLIR image is unimodality based on tremendous statistics of the gradient histogram of FLIR image. Moreover the gradient histogram of both objects and background can be fitted by χ² distribution density function with different degree of freedom. According to the above statistical characteristics of the gradient histogram and the least error-classified probability, an automatic estimation method of gradient histogram segmentation threshold is presented and validated in this paper. And the average gradient is the approximation of the optimal segmentation threshold through theoretical analysis. Finally this method is proved to be effective and practical by the segmentation experiments of the FLIR image in the real conditions.