In this paper we present a novel method for compression of focused plenoptic (light field) images. Plenoptic image is an array of lenslet images where the consecutive images act as stereo images. Adjacent lenslet images have overlapping regions leading to higher levels of redundancy than observed in a normal image. We use this overlap to appropriately select a set of reference images from which the remaining lenslet images can be approximately reconstructed. These reference images can be compressed using jpeg. The residue of the reconstructed image from the original image is compressed using either DWT-SPIHT or DCT based compression. Using the proposed method we are able to compress the image at high PSNR at a low bits per pixel (bpp).

A computer aided diagnosis (CADx) system for oral mucosal lesions has been developed using clinical cases from India as training examples. The investigated classifiers were Support Vector Machine (SVM) and Bayes Point Machine (BPM), and the task was to discriminate potentially precancerous lesions from non-precancerous lesions. The discriminating features consisted of color differences and lesions' shape properties. The overall classification accuracy was 85 % (29 out of 34) for both SVM and BPM classifiers.

No abstract

No abstract

No abstract