A Study on Pixel-wise Parallel Calculation for Depth from Focus Using Gray Level Variance

Depth-from-focus (DfF) methods can estimate the depth from a set of images taken with different focus settings. A RM method has been recently proposed and has a good performance. The method uses the relationship between the luminance value of a target pixel and the mean value of the neighbors, which follows Poisson distribution. The mean value, however, needs to be calculated using different window sizes, so an enormous amount of memory and calculation time are required. Additionally, in the calculation, box filtering to compute the mean is used twice, so a target pixel has a strong relationship with the neighbors, which makes it difficult to perform pixel-by-pixel parallel calculation for reducing the calculation time. In this paper, to address these problems, we consider modifying the image-wise calculation to the pixel-wise calculation, and propose a method that replaces the calculation with a single time filtering by approximating the formula. This can reduce the memory consumption and allows us to use pixel-wise parallel calculation.