Multiresolution representation of data in a haptic environment

A number of methods are available for the visualization of scientific data. Most of these methods use computer graphics for the visual representation of data. Such visual methods cannot be used by a blind person. Haptic interface technology makes it possible for a user to explore haptically rendered data. A haptic interface, therefore, can be used to effectively present data to a blind person. However large and complex datasets, if rendered without processing, are often confusing to the user. Additionally haptic devices are often point interaction. Thus the amount of information conveyed through the device is far less than that obtained through a visual device, making exploration difficult. Multiresolution methods provide a solution to problems that rise due to the low infomation capacity of these devices. Utilizing these methods the user can feel the data at low resolution and then add in details by increasing the resolution. These techniques are particularly useful for the visually impaired becuase complex local detail of the data often prevent the user from obtaining an overall view of the haptic plot. Wavelet is a common technique used for the generation of multiresolution data. However, the wavelet decompostion uses linear filters result in edges that are smoothed. Since nonlinear filters are known to preserve edges, we have used affine median filter in a structure similar to that used for the evaluation of wavelet coefficient. The affine median filter is a hybrid filter because its characteristics can be varied from the nonlinear median filter to a linear filter. Thus a flexible multiresolution technique with controllable characteristic is proposed. The technique is used to haptically render a 2D evenly sampled data at different resolutions. The standard Wavelet multiresolution technique is also applied to the same datasets and compared to the hybrid multiresolution technique. The advantage with the hybrid method is that with the same multiresolution structure one can go from linear wavelet decomposition to completely nonlinear multiresolution structure. It is shown that sharp edges in the orignal data are far better preserved with the nonlinear multiresolution technique. Furthermore the mean square and mean absolute errors produced in the low resolution representation are less for nonlinear techniques.