Application knowledge is available in most machine vision applications. The application knowledge can often be expressed as image properties such as shape, size, brightness, color, textures, and boundaries of image regions, edges, and lines. Application knowledge can also be expressed as geometric structures of features such as parallelism, co-linearity, orthogonality, equal-distance, adjacency, concentricity, known position, radius, etc. In industrial applications, most work-pieces have structure information available in the form of Computer Aided Design (CAD) data that specifies components as entities and blocks of entities.

Application knowledge, if correctly applied, can simplify machine vision processing and improve its outcomes. However, it is non-trivial to use the application structure knowledge in high precision applications that require sub-pixel accuracy, repeatability and real-time throughput. Conventional practice applies application structure information through a projection based averaging and edge detection approach. The projection approach integrates image pixel values in a pre-defined direction in the image to create integrated feature values that reduce noise. Edge detection is performed on the integrated feature values. This conventional approach has the following limitations:

1. The results are degraded if the image features mismatch the defined structures. For example, rotation errors result in the integration of pixel values along a wrong direction that is destructive to accuracy. In the teaching phase, human error causes mismatch of image features with defined structure. In the application phase, mismatch occurs due to imperfect repeatability of the stage or misplacement of the objects of interest.
2. The projection approach cannot effectively combine multiple structure information (such as anti-parallelism, orthogonality, intersection, or curvaceous qualities) where features of interest may be along different directions or have complex relationships. Inability to combine features limits the number of pixels involved in making a measurement and therefore decreases achievable accuracy and repeatability.
3. Resolving difficult applications requires skilled personnel to design algorithms that are purpose specific. Manual design methods produce inconsistent results and are costly and time consuming.

DRVision Technologies LLC structure-guided processing technology reads structure information directly from CAD data or through a Graphical User Interface. DRVision Technologies LLC teaching functions encode application domain structure information into processing algorithms. DRVision Technologies LLC uses the structure information to automatically enhance and detect image features of interest. The structure information compensates for severe variations such as low image contrast and noise. It retains the ability to detect true defects in the presence of severe process or sensing variations.

DRVision Technologies LLC technology automatically detects and compensates for misalignment between image features and defined structures to provide robust results.

DRVision Technologies LLC structure-guided processing technology includes a method to automatically learn processing sequence and parameters for feature enhancement. DRVision Technologies LLC structure-guided learning uses data from the application domain structure information and a target detection specification to produce a processing recipe automatically.

DRVision Technologies LLC structure guided measurement provides sub-pixel feature estimation by using structure constraints to increase accuracy. Structure constraints link multiple features to create an integrated estimation that utilizes a large number of pixels and therefore increases measurement accuracy and repeatability.