关键词: 带钢图像, 特征挖掘, 任务驱动, 图像分割, 缺陷检测

Abstract: Through the analysis of strip image and characteristics mining of defect image, this paper proposes a strip defect image segmentation method based on target characteristic mining. Firstly, this program pretreats strip defect image with median filter algorithm, then segments the strip image with a series of threshold, through characteristics mining of strip defect image and task-driven, finds out the eigenvalue mutation interval. In this mutation interval, the program segments the strip image with above method again to subdivide and finds out the optimal threshold. The program segments the strip image with the threshold and gets characteristic sub graph. Lastly, the program fuses characteristic sub images and gets the segmentation results. Experiments show that applying this method to the strip defect image segmentation process can effectively segment strip defects image. It means that this method provides a possibility for visual detection of strip defects.

Key words: strip image, characteristics mining, task-driven, image segmentation, defects detection