Restoration for ultrasonic 3-dimensional image using Boltzmann machines
Motoi Yamagiwa, M. Yoneyama · 2004
This work is concerned with the improvement method for the very degraded image obtained by the ultrasonic imaging system. In order to reconstruct 3-dimensional images from very degraded acoustic image, an auto-correlative Boltzmann Machines reformed on the visible layer is used as a post processor. The height information of the 3-dimensional image obtained from the ultrasonic imaging system consists of 12 levels. To input this type of 3-dimensional images into the Boltzmann Machines, the visible layer was reformed. The visible layer was divided into 4 divisions so that each division corresponds to input/output layer for the information of different height of 4 levels. The original 3-dimensional image consists of 16/spl times/16/spl times/12 pixels, but it is deformed to 16/spl times/16/spl times/4 pixels after the post processing by the Boltzmann Machines, because of reduction of network structure. As the result of the experiments, good recall rate of correct reconstruction can be obtained compared with that of Hopfield neural networks.