Connected operators are filtering tools that act by merging elementary regions called flat zones. Connecting operators cannot create new contours nor modify their position. Therefore, they have very good contour-preservation properties and are capable of both low-level filtering and higher-level object recognition. This article gives an overview on connected operators and their application to image and video filtering. There are two popular techniques used to create connected operators. The first one relies on a reconstruction process. The operator involves first a simplification step based on a "classical" filter and then a reconstruction process. In fact, the reconstruction can be seen as a way to create a connected version of an arbitrary operator. The simplification effect is defined and limited by the first step. The examples we show include simplification in terms of size or contrast. The second strategy to define connected operators relies on a hierarchical region-based representation of the input image, i.e., a tree, computed in an initial step. Then, the simplification is obtained by pruning the tree, and, third, the output image is constructed from the pruned tree. This article presents the most important trees that have been used to create connected operators and also discusses important families of simplification or pruning criteria. We also give a brief overview on efficient implementations of the reconstruction process and of tree construction. Finally, the possibility to define and to use nonclassical notions of connectivity is discussed and illustrated.
Guillemot, C.; Pereira, F.; Torres, L.; Ebrahimi, T.; Leonardi, R.; Ostermann, J. IEEE signal processing magazine Vol. 24, num. 5, p. 67-76 DOI: 10.1109/MSP.2007.904808 Data de publicació: 2007-10-15 Article en revista
Growing percentage of the world population now uses image and video coding technologies on a regular basis. These technologies are behind the success and quick deployment of services and products such as digital pictures, digital television, DVDs, and Internet video communications. Today's digital video coding paradigm represented by the ITU-T and MPEG standards mainly relies on a hybrid of block- based transform and interframe predictive coding approaches. In this coding framework, the encoder architecture has the task to exploit both the temporal and spatial redundancies present in the video sequence, which is a rather complex exercise. As a consequence, all standard video encoders have a much higher computational complexity than the decoder (typically five to ten times more complex), mainly due to the temporal correlation exploitation tools, notably the motion estimation process. This type of architecture is well-suited for applications where the video is encoded once and decoded many times, i.e., one-to-many topologies, such as broadcasting or video-on-demand, where the cost of the decoder is more critical than the cost of the encoder.