Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Wood Surface Inspection Using Structural and Conditional Statistical Features

Cem Ünsalan

TL;DR

The paper tackles automated wood surface defect inspection in the presence of highly textured wood patterns. It introduces support-region based features derived from gradient magnitude (GMSR) and Laplacian of Gaussian (LGSR) responses, coupled with structural and conditional statistical descriptors, and classifies defects using Bayes decision rules. The main contributions are two SR extraction methods, a set of novel structural and conditional statistical features, and a thorough evaluation showing that fused SR-based features improve defect detection and knot-type classification over traditional statistics, with promising practical impact for automated wood quality control. The findings demonstrate rotation-invariant, region-focused defect localization and discrimination, enabling more reliable automated grading of wood surfaces in manufacturing settings.

Abstract

Surface quality is an extremely important issue for wood products in the market. Although quality inspection can be made by a human expert while manufacturing, this operation is prone to errors. One possible solution may be using standard machine vision techniques to automatically detect defects on wood surfaces. Due to the random texture on wood surfaces, this solution is also not possible most of the times. Therefore, more advanced and novel machine vision techniques are needed to automatically inspect wood surfaces. In this study, we propose such a solution based on support region extraction from the gradient magnitude and the Laplacian of Gaussian response of the wood surface image. We introduce novel structural and conditional statistical features using these support regions. Then, we classify different defect types on wood surfaces using our novel features. We tested our automated wood surface inspection system on a large data set and obtained very promising results.

Wood Surface Inspection Using Structural and Conditional Statistical Features

TL;DR

The paper tackles automated wood surface defect inspection in the presence of highly textured wood patterns. It introduces support-region based features derived from gradient magnitude (GMSR) and Laplacian of Gaussian (LGSR) responses, coupled with structural and conditional statistical descriptors, and classifies defects using Bayes decision rules. The main contributions are two SR extraction methods, a set of novel structural and conditional statistical features, and a thorough evaluation showing that fused SR-based features improve defect detection and knot-type classification over traditional statistics, with promising practical impact for automated wood quality control. The findings demonstrate rotation-invariant, region-focused defect localization and discrimination, enabling more reliable automated grading of wood surfaces in manufacturing settings.

Abstract

Surface quality is an extremely important issue for wood products in the market. Although quality inspection can be made by a human expert while manufacturing, this operation is prone to errors. One possible solution may be using standard machine vision techniques to automatically detect defects on wood surfaces. Due to the random texture on wood surfaces, this solution is also not possible most of the times. Therefore, more advanced and novel machine vision techniques are needed to automatically inspect wood surfaces. In this study, we propose such a solution based on support region extraction from the gradient magnitude and the Laplacian of Gaussian response of the wood surface image. We introduce novel structural and conditional statistical features using these support regions. Then, we classify different defect types on wood surfaces using our novel features. We tested our automated wood surface inspection system on a large data set and obtained very promising results.
Paper Structure (23 sections, 14 equations, 5 figures, 6 tables)

This paper contains 23 sections, 14 equations, 5 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Support Regions
  3. Gradient Magnitude Based Support Region (GMSR) Extraction
  4. Laplacian of Gaussian Based Support Region (LGSR) Extraction
  5. Examples
  6. Support Region Representation
  7. Feature Extraction
  8. Statistical Features
  9. Structural Features
  10. Normalized Endpoint Distance ($d_n$)
  11. Compactness ($c$)
  12. Median of Curvature ($k$)
  13. Ellipse Axes Ratio Feature ($e$)
  14. Conditional Statistical Features
  15. Experimental Results
  16. ...and 8 more sections

Figures (5)

  • Figure 1: Wood surface examples and support regions extracted from them. First column: wood surface images (non-defective wood surface, wood surface with knots, wood surface with shakes and knots, wood surface with a sound and dry knot); second column: GMSR results; third column: LGSR results.
  • Figure 2: A sample wood surface image, corresponding GMSR and LGSR based curves (from left to right).
  • Figure 3: ROC curves for defect detection.
  • Figure 4: ROC curves for the knot and non-knot classification.
  • Figure 5: ROC curves for dry and sound knot classification.