Aiming at the problem of blur boundary and the intensity variation of regions in the infrared image, the traditional Maximally Stable Extremal Region (MSER) may fail to detect the region, thus leading to the poor performance. Therefore, the improved MSER algorithm is proposed in this paper to find the fault region in infrared electrical equipment image, which is based on the intensity similarity clustering. At first, the mean shift algorithm is used to cluster the pixels with similarity from the point of viewing of intensity homogeneity. Second, the thresholding mechanism is utilized to get the fast binary image, where the thresholding is selected from the low intensity of pixels clustered into the region. It thereby can split the pixels with similarity together faster, and the bright region corresponding to the fault region can be obtained through the iteration. Finally, experiments on the electrical equipment with infrared image show that the proposed method has better performance than the original MSER and owns lower misclassification error. Meanwhile, it decreases the time consumption as comparing to the Mean shift algorithm. This work is supported by Science and Technology Project of the Headquarter of State Grid Corporation of China (No. 524625160017).