Textured Surface Defect Detection (TSDD)

Computing & Wireless : Application Software

Available for non-exclusive licensing

Inventors

  • Alan Bovik, Ph.D. , Electrical and Computer Engineering
  • Sina Jahanbin , Electrical and Computer Engineering
  • Eduardo Perez , National Instruments, Inc.
  • Dinesh Nair , National Instruments, Inc.

Background/unmet need

Automatic inspection of manufactured products covered with natural-looking textures is a challenging task that is currently performed by trained human inspectors. However, the money, manpower, and time needed to perform this task are costly. A machine vision algorithm capable of detecting cosmetic or functional flaws could greatly improve the efficiency of quality control departments for a wide range of industries.

Invention Description

This software implements a computer vision algorithm that can inspect the surface of products covered with natural-looking textures for possible defects. The user first trains a one-class support vector machine to learn a general description of normal, defect-free texture in the feature space. Then switching to inspection mode, the user loads the image of the textured surface to be inspected, and the one-class support vector machine compares it to the general description of defect-free images. Finally, the location of detected defects will be displayed on the screen to the user.

Benefits/Advantages

  • Can inspect textured surfaces in real-time
  • Only requires defect-free samples for training

Features

  • The captured images are initially decomposed into several sub-bands, and these sub-band images are partitioned into non-overlapping windows.
  • The distribution of coefficient in each window is analyzed using a grey-level co-occurrence matrix and second order statistics are calculated.

Market potential/applications

Automatic Inspection, Visual Inspection, Textured Surface Quality Control

Development Stage

Lab/bench prototype

IP Status