Skip to Main Content
This paper is concerned with a technique of image compression through linear transformation which reduces the image information while generating a set of features for optimal image discrimination. This method consists of partitioning the original image into non-overlapping sub-images and applying the transgeneration technique to the subimages. The objective of this transgeneration technique is image data compression and feature generation that is optimal for image classification rather than for image representation. The technique is applied to transgenerate features fYom scintigraphic images for the detection of brain tumors. Some performance results for the classification of normal/abnormal classes of brain scans are presented. Some possible extensions and modifications of this work are briefly described.