We examine the use of high frequency features in high resolution images to increase texture classification accuracy when used in combination with lower frequency features. We used Gabor features derived from sections of 4032 2688 images. A neural network classifier was used to determine the classification performance of lower and high frequency features when used separately and then in combination. Feature shuffling and Principal Component Analysis was applied to determine both the role of each feature in the classification and to extract a smaller reduced feature set involving both lower and high frequency features.
|Translated title of the contribution||Experiments on High Resolution Images Towards Outdoor Scene Classification|
|Title of host publication||Unknown|
|Editors||Horst Wildenauer, Walter Kropatsch|
|Publisher||Vienna University of Technology|
|Pages||325 - 334|
|Number of pages||9|
|Publication status||Published - Feb 2002|