Segmentation using Texture

This paper details work on the preprocessing of phytoplankton images for classificaltion by neural networks. Firstly there is a review of the Gabor filter set generated for textural feature extraction. Secondly there is a review of discriminant analysis and the main classification techniques The third section of the repoprt details the analysis and the results of data extracted from eight test images of Dinophysis (two specimen from each of the four species). The analyses show that the background clusters well and that figure ground segmentation is possible. The analysis also shows that there is a high classification accuracy for segmenting the images upon the basis of both species and subtype (fin, theca etc.). The final section concludes that the texture segmentation using the Gabor filters is a valid preprocessing technique that will enable the group to expad the current successful neural network classifier.