Dynamic texture recognition based on complex networks GONÇALVES, W. N. ; MACHADO, B. B. ; BRUNO, O. M. International Conference on Mathematical Modeling in Physical Sciences, 2012, p. 202.
Dynamic textures have emerged as new field of investigation that extends texture images to the spatio-temporal domain. They can be defined as visual phenomena that exhibit spatial and temporal regularity. These characteristics appear in a wide range of videos, which make their analysis important in several applications of computer vision. In this paper, we propose a novel approach for dynamic texture representation based on complex networks. In the proposed approach, each pixel of the video is mapped into a node of the complex network. Initially, a regular complex network is obtained by connecting two nodes if the Euclidean distance between their related pixels is equal or less than a given radius. For each connection, a weight is defined by the difference of the pixel intensities. Given the regular network, a function is applied to remove connections whose weight is equal to or below a given threshold. Some earlier studies have shown that different thresholds reveal different texture properties of the video. Finally, a feature vector (set of measurements) is obtained by calculating the spatial and temporal average degree for networks transformed by different values of threshold. The number of connections that link pixels from the same frame and from different frames, respectively, gives the spatial and temporal degrees. Experimental results using synthetic and real dynamic textures have demonstrated the effectiveness of the proposed approach.