Browse Prior Art Database

LOCAL BLOCK SEARCH BASED AUTOMATIC DEFECT RECOGNITION IN X-RAY IMAGES

IP.com Disclosure Number: IPCOM000230618D
Publication Date: 2013-Aug-27
Document File: 9 page(s) / 1M

Publishing Venue

The IP.com Prior Art Database

Abstract

The invention proposes a template based defect detection technique for X-ray images. The technique includes computer assisted defect recognition system. The system detects defect definition/characterization which results in a more reliable inspection process. Every pixel in a test image is described by features of pixels in a block around the pixels. Further, the features are matched with a template to handle non rigid mismatch with the template. Normalized image gradients are chosen as a feature because of contrast invariance and robustness to local mis-registration. Given a test image, the technique classifies a pixel as defective or non-defective.

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 42% of the total text.

 

LOCAL BLOCK SEARCH BASED AUTOMATIC DEFECT RECOGNITION IN X RAY IMAGES

BRIEF ABSTRACT

The invention proposes a template based defect detection technique for X-ray images. The technique includes computer assisted defect recognition system. The system detects defect definition/characterization which results in a more reliable inspection process. Every pixel in a test image is described by features of pixels in a block around the pixels. Further, the features are matched with a template to handle non rigid mismatch with the template. Normalized image gradients are chosen as a feature because of contrast invariance and robustness to local mis-registration. Given a test image, the technique classifies a pixel as defective or non-defective.

KEYWORDS

Computer assisted, Template, defect recognition, pixel, non-rigid, mismatch, X-ray, amorphous.

DETAILED DESCRIPTION

X-ray radiography has become an integral part of the manufacturing process. The industrial parts once fabricated are radiographed to check for various kinds of discontinuities or flaws. An operator shifts through various X-ray images of the given part to declare it as defect-free. Any human intervention brings with it a degree of subjectivity which results in variations in repeatability and reproducibility tests.

The defects on the X-ray images are of various shapes/texture, but in many cases tend to be amorphous. Defect detection methods are classified either as reference-less or reference based. The reference-less method requires specific signatures signifying defects within the image. The reference based method is suited to detect amorphous defects as per the observed deviations from a gold standard (template).

Due to mis-registration, any pixel-by-pixel comparison with the template results in false alarms. A conventional technique known in the art includes optical flow/B-spline based image registration. However, the conventional technique is computationally expensive due to absence of specific regions of interest. Other conventional techniques include intensity based dense registration methods. However, the methods match only iso-contours of same levels without any notion of structural geometry. As a result, the conventional techniques wrap defective edges onto nearby template edges to cure the defective pixels. Figure 1 depicts curing of defects by regular dense registration. Further, spline based methods are amenable to be restricted to minimize registration metrics over a discrete set of control points. The effectiveness of such methods depends critically on the control gird placement. Moreover, a very coarse set of points do not completely capture non-rigidity and results in false alarms.

           

Therefore, there is a need in the art for a recognition system for detecting defect definition/characterization in X-ray images resulting in a more reliable inspection process.

The proposed invention describes a template based defect detection technique. Every pixel in a test...