Acquisition And Processing Of Ultrasound Images During Articulation

In this paper, we introduce the multi-modal articulation acquisition, including thehardware system and the software for controlling the hardware system by SDK.We use ultrasound machine, high-speed cameras, electroglottographs andmicrophone to acquire the articulation movement of the speaker. Because all themachines have its own sample frequency and internal clock, we design the algorithmand develop the post software to synchronize the multi-modal data precisely.Ultrasound images have serious speckle noise, which decreases the quality ofultrasound images and makes the following image-analyzing become more difficult.And we modify the standard SRAD (speckle reducing anisotropic diffusion) toimprove its edge detection and suppress the intrinsic edge blurring effect of SRAD byexploiting the multidirectional separability. We experimented with both synthetic andreal ultrasound images by SRAD and the proposed approach. The extracted contoursin denoised images by SRAD and the proposed approach are compared in terms of thecorresponding accuracy, both subjectively and objectively. The results show theproposed approach performs better than the conventional SRAD and more accuratecontours can be obtained for post processing. And we optimize the iteration numberby MSE and FOM as criteria.