“Intrackability: An Information Theoretical Measure for Good Feature to Track”
Abstract: We proposed an information theoretical measure for good feature to track, based on inferential uncertainty of video tracking. We call the inferential uncertainty intrackability, and define it as the entropy of a posterior probability. There are several classical measures for good feature to track, many of which can be regarded as approximations of intrackability with a given probability model. In addition, some theoretical justifications are given to verify that intrackability is a more general measure for feature quality. In experiments, intrackabilities are used to select not only good points to track, but also good lines to track.
“Deformable Templates for Recognition”
Abstract: Deformable template is an essential building block of the object category model. Our sketch template (i.e. Active Basis) consists of a small number of Gabor wavelet elements at selected locations and orientations. These elements are allowed to slightly perturb their locations and orientations before they are linearly combined to generate the observed image. The locations and orientations of these elements can be easily learned by a projection pursuit type of learning process.
In order to augment Active basis with texture information, the local averages of Gabor filter responses are introduced as complements of sketches. Both textures and sketches can be learned in the same pursuit framework. The learning process returns a generative model for image intensities from a relatively small number of training images. We applied the learning method to a variety of object and texture categories. The results show that both the sketches and textures are useful for classification, and they complement each other.