4.2 Combination of Dithered Data
4.2.1 Introduction to Dithering
Many ACS datasets are dithered - they have shifts between successive images which may be small or large. Dithering is a well established technique for HST imaging and has many advantages. For a general introduction to the concepts behind dithering, please consult the HST Dither Handbook (Koekemoer et al. 2002).
The primary motivations for dithering are:
- To allow the detection and flagging of image artifacts. Cosmic rays may be detected by taking multiple CR-SPLIT images at the same position, but other artifacts such as hot pixels require exposures with small dithers if they are to be detected and flagged. In addition, small-scale dithers are useful for filling in regions not covered by the detector, like the WFC chip gap or the HRC occulting finger.
- To improve the sampling of the PSF. In many ACS observation modes the pixel grid spacing is inadequate to fully sample the optical PSF falling on it. In these cases sub-pixel dithering will allow some of this lost information to be regained and the subsequent reconstruction of an improved final image.
- Larger dithers are needed to create mosaics of larger regions of sky or to cover the gap between the two chips of the ACS/WFC.
Because the non-linear component of the ACS optical distortion is large, dither shifts will vary across the field of view. This must be taken into account when planning observations. For more information on ACS dither patterns, see the ACS Patterns section of the HST Phase II proposal instructions and additional variations on the ACS "dither" webpage:
http://www.stsci.edu/hst/acs/proposing/dither.
4.2.2 Introduction to Drizzling
The methods for combining dithered ACS data which are described in this chapter are all based on the drizzle method (Fruchter & Hook, 2002 PASP, 114, 144) and are described in detail in the HST Dither Handbook (Koekemoer et al. 2002). Here we give a brief summary of the method and refer the reader to the aforementioned paper for further details and quantitative information about its properties.
Drizzle maps each pixel of the input image onto pixels in a sub-sampled output image, taking into account shifts and rotations between images and the optical distortion of the camera. However, in order to avoid re-convolving the image with the large pixel of the camera, the user may shrink the pixel before it is averaged into the output image. This shrinking factor is known as the "pixfrac" and is normally between 0.0 (corresponding to interlacing input pixels onto single output ones) and 1.0 (which corresponds to shift-and-add). The degree of sub-sampling (the size of the output pixels relative to that of the original ones) is also a user controllable parameter called "scale" which is typically in the range 1-0.5 for the ACS.
This algorithm has a number of advantages. It preserves both absolute surface and point-source photometry, hence flux can be measured using an aperture whose size is independent of position on the chip. And because the method anticipates that a given output pixel may receive no information from a given input pixel, missing data (due for instance to cosmic rays or detector defects) do not cause a substantial problem, so long as there are enough dithered images to fill in the gaps caused by these zero-weight input pixels. Finally, the linear weighting scheme is statistically optimum when inverse variance maps are used as the input weights.
The original drizzle method mapped a shrunken version of each input pixel onto the output and distributed weight according to the overlaps with the output pixel grid. For the case of multiple sub-pixel dithers of significantly undersampled images, this was close to optimum. However, for some purposes this "kernel" is not ideal and the latest implementations of drizzle allow the user to select from other options. For example, if the original method is applied to a single image, where pixfrac and scale are both normally unity, the method becomes equivalent to bilinear interpolation. This results in an output image having strongly correlated noise and also degraded resolution. For images from the ACS in the red, even the WFC channel is close to critically sampled, rather than strongly undersampled, and it would appear to be useful to consider more optimal interpolators as drizzle kernel options. Some of these are now implemented in drizzle and may also be used via the higher-level scripts described. The drizzle help file lists all the options. The "Lanczos3" kernel is particularly recommended as it minimizes apparent noise correlation in the output image and maximizes resolution. It is, however, significantly slower.
The basic drizzle method is implemented as an IRAF task within the STSDAS dither package which is invoked by PyDrizzle or MultiDrizzle.
4.2.3 Overview of Drizzling Software Tools
The tools for the combination of dithered HST images have evolved through the years, starting with those developed for the Hubble Deep Field in late 1995. First came low level tools which required considerable effort to use. Subsequently higher level wrapper scripts, written in the Python language, have been developed to ease such combination. Some of these capacities are now available within the pipeline. The main three levels applicable to current ACS imaging are as follows:
- The low level tools for mapping pixels from input to output images and support tasks for measuring shifts, detecting cosmic ray events, etc. are available within the dither package of STSDAS. These include drizzle itself and blot (reverse drizzle). The user will now probably prefer to access these tools via one of the wrappers rather than use them directly.
- PyDrizzle and associated Python scripts are layered on these and automate some of the processing. They convert the IDCTAB information into a form suitable for drizzle, create appropriate association tables and drizzle images into an appropriate output frame. These functions are implemented in the pipeline and are described in detail below. The user will again probably not use these tools directly. PyDrizzle is available within the STSDAS system and requires the PyRAF environment.
- The third level is the MultiDrizzle script which is again layered on top of the above tools. It aims to provide automated image combination, including the detection and flagging of cosmic rays. Although MultiDrizzle can be run using its default settings to do a good combination for many typical ACS dithered datasets, it also allows the user considerable control and flexibility. MultiDrizzle is available as a beta-test version but will be implemented in the STSDAS system in 2004. Like PyDrizzle, it requires the PyRAF environment.
We now briefly describe the pipeline software PyDrizzle and several associated tools. A detailed description of MultiDrizzle follows, as this is the recommended interactive tool for drizzling ACS images.
