The Journal of The Institute of Image Information and Television Engineers

A fast and memory-efficient method has been created for the dynamic mean shift(DMS) algorithm,which is an iterative mode-seeking algorithm.Running the standard DMS algorithm requires a large amount of memory because the algorithm dynamically updates all data during iterations.Therefore,it is difficult to use a conventional DMS algorithm for clustering large dataset.This difficulty is overcome by partitioning a dataset into subsets,and the resultant procedure is called a “distributed DMS algorithm”.Experimental results on image segmentation show that the distributed DMS algorithm requires less memory than that of the conventionally used DMS algorithm.

View full abstract

We previously presented a demosaicing method that simultaneously removes image blurs caused byoptical low-pass filters used in digital color cameras with Bayer's-RGB color filter array.This sharpening-demosaicing method restores only spatial frequency components lower than the Nyquist frequency corresponding to the mosaicing pattern,but it often produces ringing artifacts near color edges.To overcome this problem,we added a new total-variation regularization-based super-resolution technique.We analyzed the effects of the added technique using the DFT method.We also created a false-color suppression method using the TV regularization.Simulation results show that the TV-regularization-based super-resolution technique effectively demosaics sharp color images while restoring sharp step-edges and without producing ringing artifacts near the edges.

View full abstract

We have been investigating a multi-sensor system on which arbitrary viewpoint images can be synthesized by using image based rendering (IBR). The imaging system consists of randomly accessible smart image sensors and a field programmable gate array control circuit and generates real-time arbitrary viewpoint images. Although the system can generate more realistic images compared with the model-based rendering system, the generated images degrade if the depth information used for synthesis has errors. We have now created a method for estimating depth using the IBR imaging system. Various synthesized images are repeatedly compared with an actual image that is captured with a sensor. The depth information enables the generation of more realistic arbitrary viewpoint images.

View full abstract

Show abstractHide abstract

Using photographs that included two targets,the effects of the lens focal length on perception of the absolute distance to a single target,the relative distance between near and far targets,and the lateral distance between left and right targets were examined.Results showed that the absolute distances decreased in nearly inverse proportion to the focal length while the relative distances were barely affected by focal length though they were consistently underestimated.Because the absolute size of every target in the photographs varied in proportion to the focal length while the size ratio between two targets(relative size) did not change with the focal length,the perceived absolute distance apparently strongly depends on the target size,and the perceived relative distance is much more dependent on the relative size.Results also showed that the perceived lateral distances were approximately equal to the actual distances and not a ected by either the focal length or the target size.

View full abstract

Download PDF (4751K)

Register with J-STAGE for free!