A Breakthrough in Color, Multispectral, and Hyperspectral Based Recognition

Optical character recognition (OCR), bar coding, and related techniques have become essential tools for object recognition and tracking in modern automated manufacturing systems. However, often these tools cannot be used because of geometry, size, orientation, cost, or stage in the manufacturing process.

Since long before the invention of OCR and bar coding, humans and other living organisms have used color information for reliable, orientation independent, recognition and classification. Until recently, automation of color-based recognition has been limited to situations involving relatively simple color distributions, geometry, and lighting environment. Now new techniques, developed by the staff of Wayland Research, Inc., have made it easy to implement systems whose ability to perform complex color- based recognition is comparable with a human's, but with far greater speed and reliability.

WAY-2C is a powerful machine vision software system designed as a reliable cost-effective alternative to human labor in such tedious color-based tasks as inspection, process control, image interpretation, measurement, sorting, and yes, bar-coding. It uses Wayland Research Inc.'s powerful classification method which is uniquely suited to take advantage of the color properties of objects. Images can be from disk files, or video source such as camera, TV, or VCR. WAY-2C can be quickly trained by example to classify objects on the basis of their color distributions. It can then be placed in automatic or semi-automatic mode to perform the required task with little or no operator intervention. An easy to learn script language facilitates more complex inspection procedures. Output can be displayed on console or image monitors, written to disk file, and/or directed to serial port, printer port, or any digital output port for process monitoring or control. WAY-2C can be interfaced to databases and networks, and controlled by remote computers.

Available tools for automatic script generation facilitate dynamic reprogramming for such applications as verifying correct component sequencing for the automotive industry.

Advantages over Traditional Color-Based Systems

• Unlike simple color sensors, WAY-2C can classify many objects in different locations within a single image.

• Compared to traditional machine vision methods, WAY-2C provides dramatically improved classification while almost completely eliminating the need for color space transformations and accurate part location.

• Requirements for uniform lighting are greatly reduced often making it possible to identify objects at a distance and/or in racks or containers .

• Training by example eliminates tedious threshold, tolerance, and/or range tweaking.

• Unlike both simple color sensors and traditional color machine vision methods, WAY-2C's powerful, general statistical matching techniques make it particularly well suited for identifying multicolored objects, even those in which the proportions of the colors vary from scene to scene.

In this example distinguishing silver coins from copper coins can be particularly difficult using monochrome images because of the confusing effects of tarnish, glints, and shadows. Traditional color vision methods can also have problems. WAY-2C handles such color images with ease as shown by its interpretation portraying copper in red, silver in white and the background in black.

WAY-2C's ability to use coloring to identify objects at a distance makes it suitable for automatic identification with minimal interference with ongoing processing. For example, as long as stable lighting is installed, a WAY-2C based system has no difficulty determining which of several different color instrument panels is passing this point. Either of the regions outlined, as well as others, would be suitable for this determination. The identification information can be passed to a bar code printer, control device, plant IT system and/or simply processed by WAY-2C. With appropriate lighting entire racks of components may be inspected making WAY-2C particularly useful for automated sequence verification for the automotive industry. See also fabric recognition

In this real-time leadframe inspection application the bright red regions indicate incomplete coating covering and/or spillover. Detection of such regions generates a reject signal. Note how WAY-2C uses color information to recognize good and bad regions while disregarding details of coating shape. (Click on images to view more details).

In another typical WAY-2C electronic inspection application, the left and right images show correctly placed wires: blue, black, red and gray in order counterclockwise from the lower right. WAY-2C recognizes that the red and black wires are interchanged in the middle image, flags the offending wires, and outputs appropriate control signals.

WAY-2C easily differentiates these twisted wire pairs including the missing pair. Compare this ability to recognize multi-color combinations with the capability to recognize only single color examples offered by systems using traditional methods.

This example illustrates WAY-2C's ability to distinguish items based on subtle differences in color statistics. After being trained on the frozen dinner to the left, the system recognizes the missing meat in the upper compartment of the dinner on the right and the incorrectly placed vegetables and desert in the lower compartments.

For more WAY-2C inspection examples

Process Control

In warehouses, loading docks and similar environments, WAY-2C is used to identify and track dairy and other products packed in open crates. This information can then be passed on to control routing of the crates and/or, to accounting, billing, and labelling systems.

WAY-2C's interpretation and measure features permit determination of the location and amount of each class in an entire image or in selected regions of the image. As illustrated by this satellite image vegetation interpretation, a map (upper right) can be generated in which color codes indicate the best fitting class at each point in the original image. The map on the lower right shows regions found to be anomalous with respect to the colors in the entire original image (see Search Mode below). Click for expanded discussion

		Search mode can be used to identify regions where a particular target class is likely to be found. After training on yellow jellybeans alone,WAY-2C identifies yellow jellybean like areas in the left image by green through light gray colors in the right image.

Search mode can also be used to identify anomalous regions. After training on the left image with no yellow beans, WAY-2C finds the yellow beans in the right image to be anomalous and maps them as black.

For example, after training on unblemished product, the system detects stains and foreign objects in the first image, triggers a reject signal, and identifies the offending regions by red, orange and black in the second image.

WAY-2C's powerful statistical matching techniques make it easy to automatically determine and report the widths of the different levels of etch in this printed circuit board image. A similar approach is used for monitoring a high level nuclear waste disposal process.

WAY-2C provides information for grading and/or sorting of fruits or vegetables by either classifying based on overall color (top row), or on more detailed mapping of local color distributions (bottom row). Its unique ability to train by example and powerful statistical classification methods have been shown time and again to produce results which are remarkably consistent with those of human inspectors.

WAY-2C may be customized to perform such specialized tasks as grading and trimming asparagus. The image at the left illustrates how the system measures the curvature and thickness of the spear, the total length, and the length of the white, green, and purple sections. Typical processing speed is approximately 500 spears per minute.

The verify and search features are useful for determining whether or not an item or region falls within a specified target class. The orange region in the right image indicates an object in the left image whose colors fall outside the expected range. Green and blue regions in the right image indicate objects whose colors fall within the expected range.

Many food and beverage products use characteristic color combinations to create brand identity. WAY-2C also can use these combinations for identification. For example, it can easily differentiate the four brands of product shown at a rate of hundreds of bottles per minute, based on the colors of the cap or randomly oriented views from the sides.Training for a new brand takes under five minutes. More than twenty different brands can be differentiated at the same time. This makes it well suited for automating such applications as verifying or recording shipment information when bar-codes are unavailable or impractical.

• Assembly Inspection
• Color Matching
• Automatic Labeling
• Film Thickness
• Microscopy
• Robot Control
• Food and Materials Grading and Sorting
• Surveillance

Cost

Demonstrations, including prototype scripts for new applications can usually be arranged. Ask us about your application.