Mini-seminar on Machine Learning and HDR Imaging

School of Computing, University of Eastern Finland

26.6.2012, Conference room TB330

Learning Preference Relations with Kronecker Kernels
Tapio Pahikkala
University of Turku

In this talk, we consider a framework for learning various types of preference relations that is based on Kronecker product kernels and their modifications. As case studies, we consider tasks of inferring rankings of objects and learning to predict nonlinear preferences, as well as extensions to more complex preference learning problems. Next, we present theorems about the universal approximation properties of the considered kernel functions. Finally, we present computationally efficient learning algorithms for the considered problems and practical results on several application domains.

Anti-Ghosting in High Dynamic Range Imaging
Zijian Zhu
Institute for Infocomm Research, Singapore

Ghosting artifacts are usually caused by moving object when composing a high dynamic range image from multiple differently exposed conventional images. In this talk, a robust de-ghosting algorithm will be presented based on a double-credit intensity mapping function (IMF) and an adaptive threshold model derived from statistical training. The double-credit IMF is estimated using both pixel intensity distribution and spatial correlation. A statistical threshold model is trained from the image database, and the key parameters are determined on the fly with variance vector calculated during the IMF estimation to adapt to different scenarios. Optimal bidirectional comparison is used for further improves the detection accuracy. The experiments show the effectiveness of the proposed de-ghosting method.