New feature extraction for wood species recognition system via statistical properties of line distribution

Abstract

A key to wood identification is the distinguishable features found on the cross-sectional surface of each tree species. The surface pattern on the wood cross-section may look very similar to non-experts. However, trained experts may identify wood species based on distinct and discriminant features of the pattern. An automatic wood recognition system based on machine vision to emulate the experts, the KenalKayu has been developed with high classification accuracy. Unfortunately, when more wood species were added into the system's database, the accuracy of the system reduced. It is important for the system to have a customized feature extractor solely for wood pattern such as the statistical properties of pores distribution (SPPD) which has been proven to increase the system's accuracy. As the wood surface pattern is not only defined by pores, but lines as well, this paper presented additional new feature extraction method based on statistical properties of line distribution (SPLD) to capture the discriminant line features of each species. When used alone as feature extractor, the SPLD managed to get 88% accuracy, and the number increases to 99.5% when combined with SPPD features and 100% when combined with both SPPD and Basic Grey Level Aura Matrix features. It shows that the SPLD is an essential customized feature extractor for wood identification purposes.