Probabilistic Image Patch Recognition

Schiele and Crowley [13,12] presented a technique to determine the identity of an object in a scene using multidimensional histograms of responses of vectors of local neighborhood operators. They showed that matching such histograms can be used to determine the most probable object, independent of its position, scale and image-plane rotation. Furthermore, they showed the robustness of the approach to changes in viewpoint.

This technique has been extended to probabilistic object recognition [13] in order to determine the probability of each object in an image based only on multidimensional receptive field histograms. Experiments showed that only a relatively small portion of the image (between 15% and 30%) is needed in order to recognize 100 objects correctly. In the following we describe briefly the local characteristics and the technique used for probabilistic object recognition. The system runs at approximately 10Hz on a Silicon Graphics machine O2 using the OpenGL extension for real-time image convolution.