Locale-based object search under illumination change using chromaticity voting and elastic correlation

Searching for an object model is considered to be one of the most desirable and yet difficult searches. The problem is made difficult by the presence of clutter in a scene, as well as the fact that objects may be imaged under different lighting conditions. We have developed a feature localization scheme that finds a set of locales in an image. Our object search method matches image locales with model object locales. We make use of a diagonal model for illumination change so that each candidate assignment of model to image locales produces a possible set of lighting transformation coefficients in chromaticity space. A combinatoric search for the locale assignment problem is obviated by matching each model locale to every image locale and carrying out a voting scheme in the space of lighting coefficients. This efficiently finds the lighting change. As well, for each pair of coefficients we perform an elastic correlation on locale chromaticity. Locale centroids produce a pose estimation via a displacement model, and we can further apply texture histogram intersection and finally a generalized Hough transform efficiently since the rotation, scale and translation parameters have been recovered. Tests on a database of real images and videos show good image retrieval results