Browse Prior Art Database

AUTOMATED DEFECT RECOGNITION USING SPARSITY BASED NORMALCY MODELING

IP.com Disclosure Number: IPCOM000246116D
Publication Date: 2016-May-10

Publishing Venue

The IP.com Prior Art Database

Abstract

A technique for automated defect reconstruction in a radiographic image data is disclosed. The technique is based on sparsity based normalcy modeling. The technique includes construction of an enriched reference patch library for each pixel from a set of aligned reference images. Then, a test part is reconstructed from the reference patch library with sparsity constraints and a normalcy model is produced. Further, a residual image is calculated to represent difference between a reconstructed image and a test image. The residual image captures discrepancy between an observation and the normalcy model, thereby, highlighting defects. Furthermore, a feature based defect recognition algorithm that is tuned to recognize specific defect types recognizes and localizes the defects.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 23% of the total text.

AUTOMATED DEFECT RECOGNITION USING SPARSITY BASED NORMALCY MODELING

BACKGROUND

The present invention relates generally to radiography and more particularly to a technique for automatically identifying defects in radiographic image data corresponding to a scanned object.

A number of radiographic inspection techniques, are known in the state of the art that perform automatic radiographic inspections of scanned objects. The techniques are typically based on using assisted defect recognition (ADR) techniques to automatically screen images, call out defects and prioritize the ones that require visual inspection. ADR techniques typically achieve more accurate defect detectability than human operators and have a much higher efficiency and consistency, thereby, enhancing quality control in a wide variety of applications. Techniques using ADR may typically be used to perform automatic defect detection on two dimensional (2D) images or three dimensional (3D) images. However, such automated procedure for finding indications or defects in 2D radiography or 3D computer tomography (CT) images is a challenge due to part to part variations and imaging artifacts. Some indications or defects are very subtle and, therefore, difficult to differentiate from artifacts and part to part variations.

A conventional technique includes a method and system for identifying defects in radiographic image data corresponding to a scanned object. The method includes acquiring radiographic image data corresponding to a scanned object. The radiographic image data includes an inspection test image and a reference image corresponding to the scanned object. The method includes identifying one or more regions of interest in the reference image and aligning the inspection test image with the regions of interest identified in the reference image, to obtain a residual image. The method further includes identifying one or more defects in the inspection test image based upon the residual image and one or more defect probability values computed for one or more pixels in the residual image.

However, the above mentioned conventional technique requires explicit estimation of probability density function (pdf) or cumulative density function (cdf) and models individual pixels that does not account for differences between part to part variation and low order local registration.

Therefore, it would be desirable to have an improved technique to provide an automated defect reconstruction in radiographic image data that differentiates between part to part variation and low order local registration.

BRIEF DESCRIPTION OF DRAWINGS

Figure 1 depicts overall workflow of technique disclosed herein to determine a sample ADR result for a foreign material defect type.

Figure 2 depicts patches as yellow circle that is centered at a same pixel location and cropped through every aligned reference image.

Figure 3 depicts a view of dictionary atoms.

Figure 4 depicts a set of three sparsity, which is highlighted a...