Using radial basis function networks to approach the depth from defocus

In range finding, the depth from defocus (DFD) is a simple and effective me thod. The DFD yields the absolute depth, and does not have the image-to-ima ge matching and occlusion problems. Therefore, we use the DFD method to ana lyze the defocused images to obtain depth information using Gaussian blurre d function. In order to find the range of objects, a sigma value of the Gau ssian function due to edges out of focus is necessary. Because the sigma va lue of the Gaussian function depicts on the intensity of images grabbed by imaging devices, we employ an approximate method, the radial basis function networks (RBFN), to approach the sigma value directly in the spatial domai n. The RBFN regularizes the center position and the sigma value of the Gaus sian function to fit the profile of the defocused image by three layers of neural networks based on the radial basis function. It has accurate ranging results with less than 8% of the root mean square error in sigma value app roaching and 5% of the relative error in ranging, imaging system ranges fro m 220 min to 355 mm and focuses at 400 mm.