Statistical Object Recognition for Multi-Object Scenes with Heterogeneous Background.
Marcin Grzegorzek, Kailash N. Pasumarthy, Michael Reinhold, Heinrich Niemann · 2004
In this paper we present a statistical, appearance-based approach for localization and classification of 3-D objects in 2-D gray level images, in which the number of objects in a scene is unknown. First the statistical models of all possible object classes are created separately. The local feature vectors that we use are computed based on wavelet transformation and modeled using a normal distribution. Further, we describe a new approach for the recognition in the case of multi-object scenes. Besides the localization and classification problem, we have to estimate the number of objects in the image. For this purpose we have developed a serial search algorithm with a robust abort criterion. The experiments made on a large sample set with more than 9000 test images show that the approach is well suited for this recognition task. 1.