ISOCAM observations of the Hubble Deep Field reduced with the PRETI method

We have developed a new ISOCAM data reduction technique based on wavelet an alysis, especially designed for the detection of faint sources in mid-infra red surveys. This method, the Pattern REcognition Technique for Isocam data (PRETI) has been used to reduce the observations of the Hubble Deep Field (HDF) and flanking fields with ISOCAM at 6.75 (LW2) and 15 mu m (LW3) (Rowa n-Robinson et al. 1997). Simulations of ISOCAM data allow us to test the ph otometric accuracy and completeness of the reduction. According to these si mulations, the PRETI source list is 95% complete in the 15 mu m band at 200 mu Jy and in the 6.75 mu m band at 65 mu Jy, using detection thresholds wh ich minimize the number of false detections. We detect 49 objects in the IS O-HDF at high confidence secure level, 42 in the LW3 filter, 3 in the LW2 f ilter, and 4 in both filters. An additional, less secure, list of 100 sourc es is presented, of which 89 are detected at 15 mu m only, 7 at 6.75 mu m o nly and 4 in both filters. All ISO-HDF objects detected in the HDF itself h ave optical or infrared counterparts, except for one from the additional li st. All except one of the radio sources detected in the field by Fomalont e t al. (1997) are detected with ISOCAM. Using a precise correction for the f ield of view distortion of ISOCAM allows us to separate blended sources. Th is, together with the fact that PRETI allows to correct data on the tail of cosmic rays glitches, lead us to produce deeper source lists than previous authors. Our list of bright sources agree with those of Desert et al. (199 8) in both filters, and with those of Goldschmidt et al. (1997) in the LW3 filter, with systematic difference in photometry. Number counts derived fro m our results show an excess by a factor of 10 with respect to the predicti on of a no evolution model (Franceschini 1998) in the LW3 band. On the cont rary, the number of sources in the LW2 band is compatible with the predicti on of such a model, but with greater uncertainties, given the small number of detections.