3 Data analysis

Data were reduced using the Cam Interactive Analysis (CIA) package (Ott et al. 1998), release 3.0, jointly developed by the ESA Astrophysics Division and the ISOCAM Consortium. Dark current subtraction and calculation of the flat-field response for each pixel were done using the software provided by ESA together with the data files. The Multi-resolution Median Transform (MMT) method (Starck et al. 1996), which looks for signals on timescales shorter than the total integration time per frame, was used to identify glitches. The method proved to be robust and to work well even with unstabilized data. However, it is not designed to correct the long term gain variations of the pixel occurring after some impact of cosmic rays, because this phenomenon has a timescale longer than those examined by the MMT. This would cause spurious signals to show up after substraction of the reference image from the on-source one. To overcome this problem, the time history of each pixel after major glitches was examined and the readouts showing signs of cosmic-induced drifts were masked by hand.

It must also be taken into account that, when its illumination changes, the ISOCAM LW detector is affected by transient effects which may either cause a photometric error or the appearance of "ghost'' sources after a bright source has been observed. Corrections for transients have been computed using the method by Abergel et al. (1996; see also Starck et al. 1999).

Source fluxes were determined by means of direct aperture photometry on the maps. Fluxes were measured within radii of 6'', 9'', and 12''; the highest value was adopted. An aperture correction was applied; it was computed using the average ratio between fluxes measured in the same aperture for a theoretical point spread function (Okumura 1997) and its integrated flux on the whole array. We converted the Analogic to Digital Units (ADU) per gain and per second (ADU/G/s) of the map in mJy, using the conversion factor given with the October '98 release of CIA, that is: $1\,{\rm ADU/G/s} = 0.5068\,$mJy. In a few cases it was possible to check the flux estimate by measuring the negative imprint produced by beam-switching. Fluxes measured in this way were always in good agreement with the those measured on the positive source, even in the case of sources below the $3\sigma$ threshold. For this reason we have included in Table 2 also $2\sigma$ detections.

Astrometry of ISOCAM images, as computed from the pointing information provided by the satellite, suffers from two sources of uncertainties. The first one comes from ISO pointing accuracy. When acquiring a target, the absolute pointing error has been measured to be about $2\hbox{$.\!\!^{\prime\prime}$ }5$ ($2\sigma$) during the period when most observations have been done (Leech 1998). The second source of astrometric uncertainty is the so called "lens jitter'' of the ISOCAM instrument (Siebenmorgen et al. 1999): the position of the lens wheel in front of the detector array is not fully reproducible due to a play in the wheel. This induces a shift of the images that can reach 2 to 3 pixels (i.e. 12 to 18 arcsec) between two extreme positions of the wheel. Therefore, the astrometry of a given image is precise to 20''.

This precision is not sufficient for our purpose, but a much better astrometry can be acheived by taking advantage of the way our observations have been scheduled. The "TDTOSN'' numbers given in the last column of Table 1 consist of the revolution number when the observation was taken (first 3 digits), of the sequence number of the observation during the revolution or TDT number (next 3 digits), and of the sequence number in this sequence or OSN number (last 2 digits). Our observations were taken during 4 revolutions (398, 399, 400 and 455). During a revolution, the observations were performed on consecutive sequences of concatenated chains of beam-switch. During a concatenated chain of observations, identified by the same revolution number and TDT number, the optical configuration of the camera did not change. Hence, if we can measure the pointing offset due to the lens jitter in one of the field observed during a chain of concatenated observations, the same offset will also apply to all the observations of the chain. Between two consecutive concatenated chains of beam-switches, the camera goes back to the "standby'' mode, that is to the 6'' pixel field of view lens and to the LW2 filter. Since we are using the same lens, the lens wheel will not move between a sequence of consecutive TDT number. Therefore, if we can manage to measure the pointing offset in one of the fields of a given revolution, we can safely apply the same offset to all observations performed during that revolution.

To measure the offset we have searched for field where the optical counterpart of the ISOCAM source is identified without any doubt. Whenever possible, we have chosen fields where more than one source is detected, and where these sources are point like. Unfortunately, we only have a few fields matching these criteria, and all in revolutions 399 and 400. We have therefore used also various fields where there is only one CAM source, but with a clear-cut identification (source detected in radio by Hacking et al. (1989) or only one possible candidate). The positions of sources on the map were derived by fitting a PSF computed from the model of Okumura (1997) for high signal-to-noise sources, or using the brightest pixel center. The distortion of the map was taken into account using the measurements by Aussel et al. (1999). Positions of the optical counterparts were measured on the Digitized Sky Survey (DSS; available at stdatu.stsci.edu/dss).

For observations taken during revolution 398 we have used sources 3-09 and 3-11 having unambiguous nearby optical counterparts which are also radio sources (Hacking et al. 1989). The source in the field 3-15 would also satisfy our criteria but it is slightly extended, so that its positional uncertainty is larger. The offset between ISOCAM and optical positions is $\Delta\alpha = -12''$, $\Delta\delta = -2''$.

For observations taken during revolution 399 we have used the sources A, B and C in the field 3-19 and sources B, C, and D in the field 3-26 (source A looks extended). The derived offset is $\Delta\alpha = -21''$, $\Delta\delta = 3''$. We found a good agreement between offsets derived independently from the two fields, confirming that the lens does not move during the revolution.

For revolution 400 we used source A in the field 3-67 (source B is extended), and sources A and B in the fields 3-79 and 3-80. We obtained an offset of $\Delta\alpha = -18''$, $\Delta\delta = +3''$.

Finally, for revolution 455, we used the sources A in the fields 3-02 and 3-41 to find an offset of $\Delta\alpha = -10\hbox{$.\!\!^{\prime\prime}$ }5$, $\Delta\delta = -6''$.

The ISOCAM source positions given in Table 2 are corrected for offsets as well as for image distortions, which are particularly severe near the field edges. The contours shown in the charts appended to this paper, however, are corrected only for offsets. This is why some sources at the edges of the ISOCAM fields are not well superposed upon their obvious optical counterparts. The final astrometric precision reached after these corrections is of $3\hbox{$.\!\!^{\prime\prime}$ }25$ ($1\sigma$), $1\hbox{$.\!\!^{\prime\prime}$ }25$ coming from the absolute pointing accuracy and 3'' from the pixel size for our maps.