Three-Dimensional Shape-Structure Comparison Method for Protein Classification
IEEE/ACM Transactions on Computational Biology and Bioinformatics (TCBB)
| Hi-index | 0.00 |
This paper describes a novel application of support vector machinesand multiscale texture and color invariants to a problem inbiological oceanography: the identification of 6 species of bivalvelarvae. Our data consists of polarized color images of scallop andother bivalve larvae (between 2 and 17 days old) collected from theocean by a shipboard optical imaging system of our design. Larvaeof scallops, clams, and oysters are small (100 microns) with fewdistinguishing features when observed under standard lightmicroscopy. However, the use of polarized light with a full waveretardation plate produces a vivid color, bi-refringence pattern.The patterns display very subtle differences between species, oftennot discernable to human observers. We show that a soft-marginsupport vector machine with Gaussian RBF kernel is a gooddiscriminator on a feature set extracted from Gabor wavelettransforms and color distribution angles of each image. Byconstraining the Gabor center frequencies to be low, the resultingsystem can attain classification accuracy in excess of 90% forvertically oriented images, and in excess of 80% for randomlyoriented images.