“Selective Visualization of Anomalies in Fundus Images via Sparse and Low Rank Decomposition” by Mahurkar, Joshi, Nallapareddy and Chintala
Conference:
Type:
Entry Number: 89
Title:
- Selective Visualization of Anomalies in Fundus Images via Sparse and Low Rank Decomposition
Presenter(s)/Author(s):
Abstract:
Anomaly visualization is a well tackled problem in data visualization, especially in the context of image data. Techniques such as local contrast enhancement and histogram equalization have already been extensively explored. These techniques provide a simplistic approach to the problem, and are not geared towards selective enhancement of the anomalies. Recently, advanced segmentation algorithms have been applied, which have been shown to selectively enhance the anomalies. In spite of this, segmentation algorithms tend to be complex and ad-hoc. We provide a new approach to the problem which selectively enhances the anomalies without segmenting the regions.
In our approach, we exploit multiple images (k=10 at minimum), which combine the advantage of machine learning (where data is used constructively) with algebraic operations (rank). As such, we are able to leverage results from unsupervised data which makes the algorithm much more scalable and easier to reproduce. The dataset of healthy images need to be built only once. In contrast to standard supervised learning methods, where a dataset must be curated and the learning algorithm usually involves several tuning parameters.
References:
- Candès, E. J., Li, X., Ma, Y., and Wright, J. 2011. Robust principal component analysis? J. ACM 58, 3 (June).
- Peng, Y., Ganesh, A., Wright, J., Xu, W., and Ma, Y. 2011. Rasl: Robust alignment by sparse and low-rank decomposition for linearly correlated images. IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI) 34, 11, 2233–2246.
