Acoustic imaging is effective in extreme environments to take images without being influenced by optical properties. However, such images tend to deteriorate rapidly because acoustic impedance in air is low. It is thus necessary to restore the image of the object from a deteriorated image so that the object can be recognized in a search. We used a neural network in the previous work as a postprocessor and tried to reconstruct the original object image. However, this method needs to learn the original object image. In this work, we propose combining constructive solid geometry (CSG) with genetic programming (GP) as a new technique that does not require learning. To confirm the effectiveness of this technique, we reconstruct the image of an object from a deteriorated image created by applying a 2-dimensional sinc filter to the original image.

We developed a lips animation system that is simple and suits the characteristics of Japanese. It adopts phoneme context that is usually used in speech synthesis. It only needs a small database that contains tracing data of movement of lips for each phoneme context. Through an experiment using subjects in which the subjects read a word from lips animation, we got result as correct as from real lips.

Two kinds of discrete type Hopfield neural networks are used to recognize the facial expressions. It is expected that the associative memory effect of the Hopfield neural networks is capable of classifying the preliminarily defined specific expressions among the variety of facial expressions. Four kinds of facial expressions such as "surprise", "anger", "happiness" and "sadness" expressed by 10 persons are used as the input data to be recognized. The each image of the faces under test are divided into 8/spl times/10 regions as the feature area. Accordingly, 8/spl times/10 ternary values [+1 (moving upward), 0 (no movement), -1 (moving downward)] computed from the average value of the optical flows in each region are used as the feature parameters. Two kinds of neural networks trained by different learning data are cascade-connected to compensate each other. The experimental results showed that the averaged recognition rate for those 4 expressions was 92.2%.

By combining ultrasonic imaging with neural networks, the authors have developed a 3-D object recognition system for use in robotic vision. Ultrasonic imaging is used to calculate the initial 3-D images of the objects. These images are then passed to neural networks that identify object categories, estimate object locations, and improve 3-D images. The authors explain the 3-D ultrasonic imaging method, propose three neural network structures for 3-D image analysis, and demonstrate the practicability of the system through experimental results, which show that quick and accurate recognition can be achieved using only a small set of transducers.>