segmentation based on region growing method

 

 

 

LAN Yu jie1, ZHONG Shun cong1,2*

 

(1. Laboratory of Optics, Terahertz and Non destructive Testing & Evaluation, School of Mechanical

 

Engineering and Automation, Fuzhou University, Fuzhou 350108, China; 2. Fujian Key Laboratory of

 

Medical Instrument and Pharmaceutical Technology, Fuzhou 350000, China)

 

 

Abstract: Aiming at solving the problem of real time defect detection of eyelet fabric, an adaptive image segmentation method based on the combination of frequency domain filtering and region growing method was proposed. The power spectrum obtained from Fourier transform was analyzed, and then the yarn distribution characteristics of eyelet fabric were obtained. Specific frequency component associated with background texture was filtered and its effect on detection precision was weakened. An adaptive region growing method was employed to extract defective regions from homogeneous background in the reconstructed image. Two main parameters were automatically determined by the gray level distribution of filtered image. Morphological operations were applied to the binary image to identify defects and eventually, yield noise free images of the defects were obtained. The experimental results indicate that the adaptive and robust algorithm could localize the defect with high segmentation accuracy and therefore, it could meet the needs of industrial detection requirements.

 

Key words: eyelet fabric; defect detection; computer vision; adaptive image segmentation; frequency domain filtering