INSPECTION SOFTWARE & SYSTEMS For information about HueView, innovative inspection technology specifically developed to deliver comprehensive color analysis of multicolored, patterned and textured items, and MatchIt, a Windows-based application that facilitates its use, please click on the following link. HueView™ For information about a smart camera-based vision appliance that inspects narrow web printed pieces as well as machine marked text and graphics or labels printed on discrete items, please click on the following link: PQEye™ For information about a PC-based automated proofreading system typically used to verify the content and quality of information printed in pharmaceutical/medical product sheets and other documents that must contain accurate information, please click on the following link: PSIFEye™ For software capable of providing reliable detection of defects or important differences in images even when the images are very large, the defects small/subtle and conditions exist that make it impossible to accurately globally align the images of interest with their references, please click the following link: AVIA (Automated Visual Information Analysis)    KOJAK INSPECTOR Software For Inspecting Items Possessing Very Precise Features That Should Be Accurately Positioned Relative To Each Other Kojak Inspector uses a maximum correlation locator coupled with global image verification to automatically inspect etched, machined, precision printed and stamped parts. Parameters can even be set so as to detect subtle deterioration of the stamp itself. During an initial step, the optimal translation is determined in order to place the images of a reference or “golden” part and a unit under inspection into perfect alignment. The images are then compared point-by-point to detect differences that are larger than a user-specified threshold and as a result deemed to be defects.  A user need only supply a reference/”known good part,” the point differential that is sufficient for any region that exhibits differences to be considered a defect, and the total number of such defects that must be detected in order for a part to be rejected and possibly the press to be stopped in the case of stamping operations.  This makes the product very easy to use regardless of the complexity of the parts to be inspected.  Kojak Inspector is successfully being used in many installations worldwide, including on Yamada Machine Inc. equipment that operating at upwards of 2500 parts per minute are among the world's fastest stamping machines. The two images below show a leadframe with the kind of structural defect that Kojak detects when units under inspection are imaged using backlighting. Click to enlarge Kojak Inspector is currently available as a DLL and as one of many tools and operators that Sirius Advanced Cybernetics offers in their general purpose PC-compatible Coake machine vision solution development, test, and implementation environment.  However the most difficult aspect of automatically detecting defects in digital pictures is handling the variability present in defect-free images. Kojak Inspector or any other automated inspection solution that does not explicitly address image-to-image variation will be incapable of providing the level of performance that is possible with one that does.  Such solutions will come up short in one of two ways.  The inspection solution may be too sensitive and detect defects in areas of an image simply because of the presence of natural image variability.  Or if the solution is de-tuned to image variability, some actual defects will be missed. In situations where Kojak will not work because image-to-image variability in the form of unpredictable distortions make it impossible to achieve accurate global alignments of test and reference images, the AVIA (Automated Visual Information Analysis) inspection software will often provide the required level of performance. LOOKING FORWARD In some instances, high magnification images of items that contain precisely formed features that should be accurately positioned with respect to each other can be aligned however the visibility of normal part to part variations that would be invisible at lower magnifications precludes use of simple image subtraction as a detection technique because very large numbers of differences that are not defects would result and the post-processing required to separate true defects from natural variations that do not affect product performance would make the inspection process too inefficient. Software based on an inspection approach that effectively handles this situation is under development.  Already proven capable of reliably detecting submicron patterned semiconductor wafer defects that are of comparable size to coarse texture grains in the same field of view, this software features a "training phase" during which it examines multiple example images and then creates a statistical model of acceptable image appearance.  The statistical model allows for different amounts and types of image variation in accordance with the training samples. Whenever certain regions of the training samples contain variation that is different from other regions, this non-uniform quality of image variation is reflected in the resulting statistical model. During its "inspection phase" the software performs statistical tests on an image of a manufactured part in accordance with the “golden” statistical model developed during the training phase.  Areas of the image that are truly aberrant are flagged as defective.  Areas that merely exhibit normal image-to-image variation are not.  The software’s developers have improved its theoretically achievable level of performance by applying new techniques to the underlying algorithms and making it possible for sensitivity and specificity to continuously improve as more and more images are inspected.  Stay tuned for the introduction of a C/C++ library that implements the improved algorithms.