Fourth issue of real-time image processing in 2009

The constantly increasing number of submissions to the Journal of Real-Time Image Processing (JRTIP) reflects the growing importance of the real-time demands in many image processing areas where a few years ago not many thought that real-time performance could ever be achieved. Solutions comprise all aspects for boosting processing speed by hardware and software implementations that can be generalized and ported to new applications. The editors-in-chief of JRTIP have considered this increasing interest in the field by providing a platform for exchange of related ideas and developments with this journal through the SPIE annual conference on Real-time Image and Video Processing. In 2010, this conference will take place for the first time in Europe. The current issue of JRTIP concludes four successful years of publishing papers solely dedicated to the real-time aspects of image and video processing. The plan is for volume 5 in 2010 to include two regular and two special issues, the next one of which is scheduled to be on ‘‘Improving Display and Rendering Technologies for Virtual Environments’’. This fourth issue of the fourth volume comprises five original research articles, all of them dealing with motion in one way or another addressing several aspects of object recognition in video sequences ranging from object extraction and tracking to an application in quality control for food industry. Following the trend of retrieving image information directly in the compressed domain, the first paper demonstrates a video object extraction for MPEG-2 video stream with real-time performance where a dense and accurate motion vector field is generated based on a confidence measure. The second paper presents a segmentation of outlines of moving objects for interactive vision systems in real-time, which is implemented and tested on a dedicated VR platform. The third paper introduces a two-stage motion prediction as a mechanism to work around the major bottleneck associated with the speed of video encoding, i.e. the computational complexity of the bidirectional motion estimation (ME) in motion compensated temporal filtering due to filter size. This combination of zero motion block prediction and inter-frame and inter-layer prediction techniques generates results with negligible decrease in quality. The fourth paper contributes to the field of Systemon-Chip (SoC) architectures by a reconfigurable design consisting of a combination of microprocessor and FPGA. Within a Belief-Desire-Intention based multi-agent architecture (BDI), one agent-based task graph and two heuristic partitioning methods are proposed to partition the hardware and software on a reconfigurable SoC platform. Beside efficiency and robustness, real-time performance of object tracking is demonstrated implementing a particle swarm optimization for object detection and tracking. Finally, the last paper is dedicated to a fast sorting application in food industry implemented in an efficient and reliable rice-sorting machine working in high speed with high classification accuracy. The article focuses on the selection of a subset of Zernike moments and an application of a multi-layer perceptron. A combination of fuzzy ARTMAP and genetic algorithm speed up the computation time for selecting the subset of Zernike moments considerably thus maintaining the classification accuracy at the same time. M. F. Carlsohn (&) Bremen, Germany e-mail: matthias.carlsohn@t-online.de