Cosine Lobe Based Relighting from Gradient Illumination Photographs

We present an image-based method for relighting a scene by analytically fitting cosine lobes to the reflectance function at each pixel, based on gradient illumination photographs. Realistic relighting results for many materials are obtained using a single per-pixel cosine lobe obtained from just two color photographs: one under uniform white illumination and the other under colored gradient illumination. For materials with wavelength-dependent scattering, a better fit can be obtained using independent cosine lobes for the red, green, and blue channels, obtained from three achromatic gradient illumination conditions instead of the colored gradient condition. We explore two cosine lobe reflectance functions, both of which allow an analytic fit to the gradient conditions. One is non-zero over half the sphere of lighting directions, which works well for diffuse and specular materials, but fails for materials with broader scattering such as fur. The other is non-zero everywhere, which works well for broadly scattering materials and still produces visually plausible results for diffuse and specular materials. We also perform an approximate diffuse/specular separation of the reflectance, and estimate scene geometry from the recovered photometric normals to produce hard shadows cast by the geometry, while still reconstructing the input photographs exactly.

Content Classification of Multimedia Documents using Partitions of Low-Level Features

Audio-visual documents obtained from German TV news are classified according to the IPTC topic categorization scheme. To this end usual text classification techniques are adapted to speech, video, and non-speech audio. For each of the three modalities word analogues are generated: sequences of syllables for speech, “video words” based on low level color features (color moments, color correlogram and color wavelet), and “audio words” based on low-level spectral features (spectral envelope and spectral flatness) for non-speech audio. Such audio and video words provide a means to represent the different modalities in a uniform way. The frequencies of the word analogues represent audio-visual documents: the standard bag-of-words approach. Support vector machines are used for supervised classification in a 1 vs. n setting. Classification based on speech outperforms all other single modalities. Combining speech with non-speech audio improves classification. Classification is further improved by supplementing speech and non-speech audio with video words. Optimal F-scores range between 62% and 94% corresponding to 50% - 84% above chance. The optimal combination of modalities depends on the category to be recognized. The construction of audio and video words from low-level features provide a good basis for the integration of speech, non-speech audio and video.