Active contour based rock sole recognition

Abstract

In this paper, we describe a system for classifying rock sole fish. The contour of the fish and existence of a stripe pattern constitute two important features to distinguish between different subspecies of rock sole. The system uses a multiscale sigma filter and active contour algorithm to extract the contour of the groundfish. The extracted contour is then transformed into a one dimensional signature to serve as the first feature to differentiate the northern rock sole from the southern rock sole. Next, our system uses a pattern enhancing algorithm and transforms the extracted pattern to the spatial frequency domain to serve as the second feature. A matched subspace filter is then applied to recognize each feature. The proposed recognition system integrates the classification result from both channels. We illustrate the performance of the procedure on sample images of northern and southern rock soles.