| | 14 | After the initial input rejection, a target PSF is necessary to know how much each input must be convolved to match the other inputs. Once this target has been identified, individual kernels are constructed for each input warp to convolve those images to the target. This step is essential to ensure that the pixel level rejection operates on images that have the same object shapes. Following this, a source list is compiled by finding the matched set of objects on the individual frames, with magnitudes calculated from the average of all images that have detections of that source. |
| | 15 | |
| | 16 | For PV1, the target PSF was constructed as the "envelope" of the set of input warps. Briefly, this process attempts to construct a PSF that approximately the maximum of each inputs. Once this target has been constructed, each input image is matched to a target frame constructed by simulating the object catalog with the target PSF using the Alard-Lupton algorithm. The kernels for this case are ISIS kernels, a sum of Gaussians and polynomial terms. PV1 had some difficulties in this matching process. First, the envelope PSF could be derived with a width larger than that of any input image PSF. This was not ideal, as it increased the level of smoothing present on the final image. Secondly, the Alard-Lupton process with ISIS kernels can construct odd weightings for individual frames, and the large number of parameters ... czw: I can't remember all the issues. |
| | 17 | |
| | 18 | PV2 has attempted to solve these problems by using less complex methods. The target PSF is now constructed to be a single Gaussian with FWHM equal to that of the largest input PSF. This FWHM can be incremented by a small amount (currently 0.1 pixel) to ensure that no input needs to be deconvolved. With the target PSF represented by a single Gaussian, the convolution kernels for all input warps are defined to also be Gaussians, with FWHM_kern values equal to the value that would be required if the input PSF were a Gaussian with the same FWHM. Therefore (FWHM_kern)^2^ = (FWHM_target)^2^ - (FWHM_input)^2^. |
| | 19 | |
| | 20 | In both PV runs, after determining the kernels, the input image/mask/variance images are convolved by the appropriate kernel. |
