Prediction of the grey level intensity in selected windows of image sequence using radial basis network

Abstract

An experimental study of the turbulent mixing flow in the wake of a prismatic bluff body was made in a nonreturn subsonic wind tunnel (Re/sub b/=4300) using flow visualization and a digital image-processing technique. A high-speed camera was used to capture smoke visualization images of the turbulent mixing flow structures. From the grey level intensity of selected image windows, using a radial basis neural network, grey levels of the neighbouring locations were calculated and compared to the measured intensity. As an input area, part of the image was used, located upstream of the prediction area. Prediction was based on history of 6 successive images. Neural network was trained where first 200 images of the same sequence were applied. Quality of prediction depends on now properties at a given location and on the distance from the input area. The quality of prediction at various locations corresponds well to the intensity of concentration fluctuations. Power spectra of the predicted and actual image sequence are compared.