Deriving a catalog of discrete sources from astronomical images is not a new problem and a number of well-documented computer programs already exist for this purpose (e.g., DAOPHOT). However, most of these programs were developed specifically for extracting point sources from optical images. They are generally not well suited for use with high-resolution radio images in which a significant fraction of the sources are clearly extended. In addition, the noise in radio images has peculiar characteristics that cause trouble when the optical source-finding procedures are used (e.g., it varies from pixel-to-pixel and has some large-scale residual structures remaining after the CLEANing is complete). Furthermore, since all of the images in the FIRST survey are generated in AIPS, we considered it essential to have an AIPS-based source extraction system. The 15JUL93 release of AIPS contained a source extraction program named SAD (for Search and Destroy) which included a number of serious errors, but which served as a good starting point for our own efforts to create a functional AIPS-based routine for source extraction. We call the new task which resulted from our efforts HAPPY (for how we feel about this survey.)
HAPPY starts by searching an image for pixels which exceed a user-provided threshold. It then defines the minimum-size rectangle around each contiguous set of threshold-exceeding pixels. The resulting rectangles, or `islands', are padded with a border 3 pixels wide. If this border touches another island, any border-strip pixels exceeding the threshold are ignored in the subsequent fit. The pixels in each island are searched for local maxima, which are used as initial estimates for a two-dimensional Gaussian fitting algorithm (similar to the AIPS task JMFIT). Up to four Gaussian components can be fitted to each island. The program is permitted to find up to 10,000 islands per image, although no more than the brightest 2000 islands are actually fit. A typical FIRST survey coadded image has 35 islands.
The individual islands are analyzed in order of decreasing brightness. As the components in each island are fit, the results of the fitting algorithm must pass several acceptance criteria:
If any of these criteria are not met, the island is refit with one fewer component. Islands are also rejected if they intersect the edge of the image; such areas of sky are always covered more completely in the adjacent image. The results of a successful fit are subtracted from the original image before the next island is fit.
The end product of HAPPY is a list of elliptical Gaussian components. For each component HAPPY gives an RA and Dec, peak and integrated flux densities, major and minor axes, and the position angle of the major axis measured east from north. In addition, an error estimate is calculated for each parameter, and an island number (where island 1 contains the brightest pixel in the field, island 2 the second brightest non-contiguous pixel, etc.) and the number of components within each island are recorded. The peak and integrated flux density for the island are also saved; these quantities may be useful for identifying complex islands that are poorly modeled by the Gaussian components. Some of this information is not included in the currently distributed version of the catalog, but it has not been discarded and may be made available in future versions.
The position angles in the catalog have a somewhat peculiar distribution. Sources that are not detectably elliptical (through a combination of circular symmetry and faintness) have position angles concentrated near 0°, 90°, and 180°. These ``magic values'' are determined by the initial guess for the position angle for each source and do not have any special significance.