Significant effect of image contrast enhancement on weld defect detection

Area of Science:

• Industrial NDT (Non-Destructive Testing)
• Computer Vision
• Image Processing

Background:

• Manual weld defect inspection is prone to errors and uses outdated X-radiography methods.
• Digital X-radiography images require advanced processing for accurate defect detection.
• Contrast variation and luminosity issues hinder image quality and segmentation performance.

Purpose of the Study:

• To propose a novel Hybrid Statistical Enhancement (HSE) method for weld defect inspection.
• To improve image quality by addressing contrast variation and luminosity.
• To enhance the performance of automated weld defect detection and classification systems.

Main Methods:

• The HSE method utilizes statistical data (mean and standard deviation) of pixel neighborhoods.
• Pixels are classified into foreground, border, and problematic regions based on luminosity and contrast.
• Weld defect images were processed using HSE and segmented with Bernsen's and Otsu's methods.

Main Results:

• HSE effectively enhances contrast variation and normalizes luminosity in weld defect images.
• Segmentation results using HSE showed lower Misclassification Error (ME) compared to Homomorphic Filter (HF) and Difference of Gaussian (DoG).
• Accuracy in weld defect detection increased from 64.171% to 84.964% after applying HSE and segmentation.

Conclusions:

• The proposed HSE method offers an effective and efficient solution for background correction and image quality improvement in weld inspection.
• HSE facilitates automated weld defect detection and classification, leading to more reliable industrial testing.
• The approach demonstrates significant improvements in segmentation and detection accuracy for weld defect images.