The method that PhabrOmeter used has been a common practice for many years by ladies in certain parts of the world when searching for a desired scarf at a market. They would take off their rings and pull out a scarf through the ring, judging the overall quality of the scarf based on the resistance during the pulling out process.

PhabrOmeter simulates such method by extracting a fabric sample through a properly designed nozzle. During the process the sample is deformed under a very complex yet low stress state including tensile, shearing and bending as well as frictional actions, similar to the stress state when it is handled by people. Consequently, all the information related to fabric hand is reflected by the resulting load–displacement extraction curve.

Once the fabric extraction is done and an extraction curve is obtained, several following tasks are to be accomplished :

• To reduce the dimensionality of the data set

  The fact that the extraction curve is a continuous curve implies a problem: all the points on the curve are more or less correlated, as we can predict where a given data point will be located by knowing the locations of its prior and/or subsequent points; the closer the neighboring points, the higher the correlations between them.

  These ominously existing correlations among the data indicate a redundancy of information contained in the data points. Therefore, for both simplicity and facilitation of further analysis, we have to eliminate the redundancy and consequently reduce the number, or the dimensionality, of the data set.

• To identify new and meaningful variables

  With fewer but condensed data points or features retained, we need to determine their identity or physical meanings so as to interpret the results. For different fabrics, there are different extraction curves, and detection and quantification of the differences, determination of the physical significance of the features, are just what the pattern recognition techniques deal with.

By testing fabrics using the extraction method, the extraction curve (displacement vs. extraction force) is discredited into a data set X which is then transformed into the feature vector Y. Based on physical calibration and numerous validation, the first three components /features of the Y set are highly correlated with the fabric attributes, Resilience, Softness & Smoothness, and in that order. So far, we provide to users only the first three features in the set Y and the remaining five are only included in calculating an overall Relative Hand value described below, until practical needs for more fabric hand features thus defined arise.

These features: Resilience, Softness & Smoothness are mathematically determined independent parameters, whereas the actual such fabric hand attributes are from human sensory responses that are inevitably intertwined with each other. It is therefore illogical to attempt associating these calculated features with the sensory attributes, no less than associating individual primary colors with the hue or tune of the resulting color perception.

However, to make our system easier to be accepted, we have spent a huge amount of time to establish some connection between two groups that define the same fabric hand, albeit through different paths. After all, these sensory attribute terms (and their equivalences in other languages) have been used by people for so long in describing fabric hand.