This is an online tool that performs deconvolution using the MODIFIED CLEAN algorithm. This algorithm is quantitive (i.e., deconvolved values are on the same scale as the input data) and conservative (i.e., the flux of the residual plus the flux of the clean map equals the flux of the original image) making it suitable for use where photometic stability is important.
This tool uses TIFF files for input and output. TIFF is optimised for image transport, not for describing the units the image is in. Therefore, the deconvolved results have their range (in terms of the input image range) added to their filename when they are downloaded.
Currently it has been tested to cope with single-colour or 3-colour data in (un)signed 8 or 16 bit integers. Each colour is deconvolved as an independent layer. Therefore a science image and a PSF must either 1) have the same number of colour layers or 2) the PSF has one colour layer that is applied to all the science image colour layers. After deconvolution, the colour layers are recombined in the same format as the input science image.
Note: While the algorithm used conserves flux, the TIFF format does not know about flux. Therefore we use the entire available range of values.
Residual
Figure 1: Shows the residual (the components at the current iteration convolved with the PSF, subtracted from the original image) at the current iteration.
This should progress from being identical to the original image at iteration 0, to being mostly (if not entirely) noise at the final iteration.
Selected Pixels
Figure 2: Shows the selected pixels at the current iteration. Pixels are selected as those whose magnitude is larger than the brightest pixel in the residual multipled by the threshold of the current iteration. At each iteration the selected pixels are multipled by the loop_gain and added to the current components. This image gives direct insight into what is happening each step and you are often able to pinpoint the stage in a deconvolution where things start to go wrong by watching this image.
When using adaptive_threshold_flag this may look similar to the residual (especially at the start) as adaptive thresholding attempts to initially separate the image into low brightness "background" and high brightness"foreground", so in the vast majority of images the object will be the "foreground" initially. As iterations progress, bright features start to be classified as "foreground".
When using a manual threshold, this will often look very different from the residual at all times. This is because a manual threshold is not dynamic, generally the selected pixels will start of as a contiguous blob, but become more fragmented over time. This can be a source of speckling that degrades the final image, therefore adaptive thresholding is recommended.
Current Convolved
Figure 3: Shows the currently selected pixels convolved with the PSF. This looks like (and is) a "smoothed out" view of the selected pixels. You can think of this image as what will be subtracted from the current residual to get the next iteration's residual.
Current Components
Figure 4: Shows the current components of the deconvolved image. If deconvolution were to stop, this would be come the clean map (if no clean beam is used and the residual is not added).