|
|
XCaliper 6.0 |
|
Products & Services
|
NoisemasterTM Robust Pattern Technology The Noisemaster pattern-recognition tools in XCaliper 6.0 provide new algorithms for pattern-recognition tasks where images are of poor contrast, degraded with occlusions, or generally noisy and inconsistent in appearance. These capabilities are similar to those promoted by geometric pattern tools but add the benefit of handling patterns of objects whose edges are not well-defined. Noisemaster algorithms have been optimized to automatically take advantage of systems with dual processors, so they are ideal for the most demanding applications where search times may be critical.
When To Use The Standard NFT TeachPattern and TeachFont Tools? We recommend using the existing NFT-based tools as a first approach to any application where image quality appears fairly consistent and is not prone to degradation from major shifts in illumination or reflection or from occlusions introduced by the manufacturing process. Since the more robust pattern recognition in the new tools requires more intensitve computations, the NFT tools are somewhat faster. NFT tools are also the better choice for applications in which the ability to make fine distinctions in patterns is the dominant criteria. Take the simple example of distinguishing between two very similar objects such as the letter “Q” and the letter “O”. Given a partial occlusion of the tail on the letter Q, NFT tools are more likely to make the distinction than Noisemaster. While NFT tools make finer distinctions, Noisemaster is better able to recognize patterns in degraded images. When To Use The New Noisemaster RPT-TeachPattern and RPT-TeachFont Tools The new RPT tools should be applied to any application which you expect to be analyzing noisy images or images in which the patterns themselves have inconsistent shapes or definition. The task of consistently finding the poorly-defined fiducial marks in Figure 1 is a good example of such conditions from a real world machine vision application. These fiducials present the dual problem of very irregular object edges and minimal contrast with the background substrate. (Note the low contrast shown in the intensity profile generated from the line transect across the rightmost mark.)
Figure 1. Fiducial marks with irregular edges and poor contrast Figures 2 through 6 demonstrate the specific differences in Search results using classifiers created with both NFT and RPT. Classifiers were created using the same parameters in both tools as demonstrated in Figure 2. |
|
|
|
Figure 5. Search results on occluded fiducial
marks using RPT tools
(Left: Modified image showing significant occlusion of original patterns. Right:
Search results)
Figure 6 demonstrates the ability of the new Noisemaster tools to accommodate patterns with complete contrast reversal. This feature is provided by the “invert” variant option, which can be applied during the training process.
Figure 6. Inverted contrast patterns found with RPT tools
Can Both Technologies Be Used Together?
| Given the unique capabilities of both sets of pattern recognition tools in XCaliper 6.0, a commonly- asked question is if both types of tools can be used in the same application. The short answer to this question is “Yes.” as demonstrated in the new sample application (RPTSrcTch) included in the XCaliper 6.0 install. | As with all of the other tools in the XCaliper toolset, any combination of classifiers may be combined in the runtime application. For example, the application could have a single Search tool cycle through a list of classifiers and then re-execute a single search. Or, it could create multiple Search tools using different classifiers and execute them separately in order of importance. |
Download Noisemaster Application Note PDF.
|
Copyright © FSI Automation, Inc. 2004 |
