JOURNAL OF APPLIED COMPUTER SCIENCE Vol. 18 No. 1 (2010), pp. 87–115 Spatch Based Active Partitions with Linguistically Formulated Energy Arkadiusz Tomczyk, Michal Pryczek, Stanislaw Walczak, Konrad Jojczyk, Piotr S. Szczepaniak Institute of Information Technology Technical University of Lodz Wólcza´ nska 215, 90-924 Lód´ z, Poland tomczyk@ics.p.lodz.pl Abstract. The present paper shows the method of cognitive hierarchical ac- tive partitions that can be applied to creation of automatic image under- standing systems. The approach, which stems from active contours tech- niques, allows one to use not only the knowledge contained in an image, but also any additional expert knowledge. Special emphasis is put on the efficient way of knowledge retrieval, which could minimise the necessity to render information expressed in a natural language into a description con- venient for recognition algorithms and machine learning. Keywords: image understanding, image segmentation, active partitions, lin- guistic. 1. Introduction In the traditional approach to image analysis, an image is first preprocessed (e.g. by means of noise elimination or edge retrieval) and then analysed by means of segmentation, which employs such techniques as thresholding, region growing and splitting, etc. Such approaches are described in [1, 2, 3]. Most of them use only the knowledge contained in the image. However, researchers e.g. in [4] have started