V. Praneeth (India)
Feature extraction, Texture, Spectrum, Image classifica tion, ILHS space, Triangle inequality
In this paper, an orthogonal subspace projection (OSP) ap proach, successfully used on hyperspectral images has been chosen for classification of image data prior to segmenta tion as it has a distinct advantage of separating undesired signatures and interferences within a pixel by decomposing a pixel into a signature space and a noise space, where the desired signature is retrieved using a matched filter. Now our main problem, prior to using an OSP approach is that an OSP requires a good knowledge of signature abundances and expressive features and characteristics of images to be collected for an OSP based classifier to function properly. Hence, it is required to efficiently extract feature vectors for training purposes from images. We have chosen to ex tract feature vectors based on the important image char acteristics of ’texture’ and ’colour’. The two procedures: (a.)Texture Spectrum and (b.)Colour Histogram from ILHS 3D-Polar coordinate space are used for extraction of feature vectors from the given image. Finally, various criteria for comparison of images based on texture spectrum have been mentioned. An advantageous application of such measures is that, they can be used for developing algorithms for ef ficient content based retrieval from image databases. Im plementation of this approach presented has been done in ’Matlab’ and the results of the texture spectrum, ILHS his togram verified with the spectrum given in [2] before pro ceeding to images of our own interest.
Important Links:
Go Back
