A Co-Random Walks Segmentation Method for Aerial Insulator Video Images

Insulator segmentation is an important premise of automatic state detection and fault diagnosis in image processing. The aerial insulator images are characterized by complex background, low resolution, large number and many pseudo targets. The classical random walks algorithm may segment wrongly. It requires massive interaction to segment multiple images, which makes user fatigue and results in bad segmentation quality. This paper proposes an automatic co-segmentation method called co-random walks, which utilizes the relationship between aerial insulator video images as prior information to find corresponding seed points in order to achieve higher segmentation accuracy. Firstly, we remove the texts in original images, the preprocessed images are over-segmented into super-pixels by SLIC (Simple Linear Iterative Clustering) for fast segmentation. Then the collaborative graph network is constructed, we use a greedy algorithm to get corresponding seed points. Finally, the random walks segmentation for corresponding seed points of each image is performed. The experimental results show that the proposed method can efficiently distinguish the insulator from complex background and eliminate the pseudo target like tower. We only need a few seed points to achieve relatively high accuracy of automatic segmentation, which is instrumental to unmanned aerial vehiclel aerial insulators' state detection and fault diagnosis.