Edge Detection Techniques for Automatic Location of Spectra

To improve the processing of multi-object or cross-dispersed spectroscopic data, especially for systems resulting in curved 2-D spectra, we have implemented in Python edge detection techniques widely used in the photo processing and remote sensing world. The software uses the discontinuity found in a spectral image to precisely locate each dispersed 2-D spectrum on the pixel array. A valid spectrum image edge is defined as continuous and sharp. To this end the best input data is a well illuminated flat field. The algorithm applies a discontinuity detection filter to the image. We find that a 3 x 3 Sobel kernel reliably produces easily traceable edges on our data. Some instruments produce data with large background noise. In those cases, a mild smoothing filter is first applied to reduce noise spikes that would otherwise confuse the edge tracing algorithm. The edges highlighted by the filtering are traced using the SciPy function label. Each edge is represented by a second degree polynomial that follows each slit edge. Currently the software assumes that the spectra are nearly horizontal or nearly vertical. This constraint can easily be lifted with the choice of a different convolution kernel.