Machine vision is increasingly used to provide automated situational awareness, for factory automation, robot device guidance and many other applications.
Whether it is for a simple Pass/Fail test, determining which one out of hundreds of fabrics is present, quickly recognizing an anomalous situation, or determining the nature of some potential hazard, this usually involves some type of classification.
We humans typically first define a set of classes of interest. We then use our observations to decide to which class the current situation is most likely to apply. In automation, these observations are generally measurements of one or more attributes of the item to be identified. The greater the consistent contrast in these attributes among the classes, the more certain we can be that our classification will be correct.
Well-established mathematical methods allow us to quantify such terms as contrast and the probability of correct classification. However, the general principles are qualitatively familiar to everyone.
Generally speaking we seek a set of attributes that makes it easiest to identify a class or anomaly. Good contrast among the attribute values for the classes of interest is key to obtaining speed and certainty in the classification. We have developed and patented a method that, when many different attributes are available, allows one to determine an optimum subset combination for maximum contrast.
In the field of vision, color is one way that contrast is increased. Color is a perception resulting from the combination if light signals from a single origin passing through several sets of filters in the eye and combined in the brain. As the number of filter sets increases so does the potential for increased color contrast. Most humans have three filter sets; some have fewer; a few women have four sets. Color cameras provide the same type of filtering making signals available for automated processing.
When items of interest have distinct color combinations, contrast for identification purposes is further increased without the need for additional filter sets, sharper color resolution or greater consistency of the individual colors. Thus color complexity can result in more reliable classification than that based on a single color.
Unlike traditional color machine vision software, WAY‑2C is built on a solid foundation of the laws of modern probability and information theory to take advantage of the enhanced contrast available from both color and color complexity. This allows it to be quickly, easily, and automatically trained to achieve robust, reliable, color-based identification almost anywhere a human can. The laws apply to any color, color-like or "false-color" image based on information from virtually any source data combination. Therefore WAY‑2C and its methods are equally applicable to all such images.