Roughness preserving filter design to remove spatial noise from stereoscopic skin images for stable haptic rendering.

BACKGROUND/PURPOSE: A problem in skin rendering with haptic feedback is the reconstruction of accurate 3D skin surfaces from stereo skin images to be used for touch interactions. This problem also encompasses the issue of how to accurately remove haptic spatial noise caused by the construction of disparity maps from stereo skin images, while minimizing the loss of the original skin roughness for cloning real tough textures without errors. Since the haptic device is very sensitive to high frequencies, even small amounts of noise can cause serious system errors including mechanical oscillations and unexpected exerting forces. Therefore, there is a need to develop a noise removal algorithm that preserves haptic roughness.

METHODS: A new algorithm for a roughness preserving filter (RPF) that adaptively removes spatial noise, is proposed. The algorithm uses the disparity control parameter (λ) and noise control parameter (k), obtained from singular value decomposition of a disparity map. The parameter k determines the amount of noise to be removed, and the optimum value of k is automatically chosen based on a threshold of gradient angles of roughness (Ra ).

RESULTS: The RPF algorithm was implemented and verified with three real skin images. Evaluation criteria include preserved roughness quality and removed noise. Mean squared error (MSE), peak signal to noise ratio (PSNR), and objective roughness measures Ra and Rq were used for evaluation, and the results were compared against a median filter. The results show that the proposed RPF algorithm is a promising technology for removing noise and retaining maximized roughness, which guarantees stable haptic rendering for skin roughness.

CONCLUSION: The proposed RPF is a promising technology because it allows for any stereo image to be filtered without the risk of losing the original roughness. In addition, the algorithm runs automatically for any given stereo skin image with relation to the disparity parameter λ, and the roughness parameters Ra or Rq are given priority. Although this method has been optimized by graph-cut disparity map building, it can be extended to other disparity map building methods because the parameter k is determined by actual roughness Ra data that can be obtained by simple measurement.