1999A&A...342..363D

We have developed a general data reduction technique for ISOCAM data coming from various deep surveys of faint galaxies. In order to reach the fundamental limits of the camera (due to the background photon noise and the readout noise), we have devised several steps in the reduction processes that transform the raw data to a sky map which is then used for point source and sligthly extended source extraction. The main difficulties with ISOCAM data are the long-term glitches and transient effects which can lead to false source detections or large photometric inaccuracies. In many instances, redundancy is the only way towards clear source count statistics. A sky pixel must have been ``seen'' by many different CAM pixels. Our method is based on least-squares fits to temporal data streams in order to remove the various instrumental effects. Projection onto the sky of the result of a ``triple-beam method'' (ON-(OFF1+OFF2)/2) obtained from the signal of a given pixel during three consecutive raster positions leads to the removal of the low frequency noise. This is the classical approach when dealing with faint sources on top of a high background. We show illustrative examples of our present scheme by using data taken from the publicly available Hubble Deep Field ISOCAM survey in order to demonstrate its characteristics. More than thirty sources down to the 60 (resp. 100)µJy level are clearly detected above 4 σ at wavelengths of 7 (resp. 15) µm, the vast majority at 15µm. A large fraction of these sources can be identified with visible objects of median magnitude 22 and K-band magnitude of 17.5 and redshifts between 0.5 and 1 (when available). A few very red objects could be at larger redshifts.