EigenBlock algorithm for change detection - An application of adaptive dictionary learning techniques

Change detection methods are very important in many areas such as medical imaging and remote sensing. In particular, identifying the changes in medical images taken at different times is of great relevance in clinical practice. The key of detecting changes in medical images is to detect disease-related changes while rejecting “unimportant” induced by noise, mis-alignment changes, and other common acquisition-related artifacts (such as inhomogeneity). In this paper we first summarize the existing methods for automatic change detection, and propose a new approach for detecting changes based on local dictionary learning techniques. In addition we aim to automatically ignore insignificant changes. Our new approach uses L2 norm as similarity measure to learn the dictionary. We also apply knowledge of principal component analysis as a feature extraction tool, to eliminate the redundancy and hence to increase the computational efficiency. The performance of the algorithm is validated with synthetic and clinical images.